Embed Size (px)
Citation preview
DOCUMENT RESUME
ED 081 447 LI 004 455
AUTHOR Wilman, H.; Hall, Angela M.TITLE INSPEC: An Experiment in the Batch Processing of
Retrospective Searches.INSTITUTION Institution of Electrical Engineers, London
(England).SPONS AGENCY Office for Scientific and Technical Information,
London (England).REPORT NO INSPEC-R-73-13PUB DATE Feb 73NOTE 41p.AVAILABLE FROM INSPEC, The Institution of Electrical Engineers,
Savoy Place, London WC2R OBL, England (pound 1.25)
EDRS PRICE mF-$0.65 HC Not Available from EDRS.DESCRIPTORS Automation; Comparative Analysis; Computers; Costs;
Information Processing; *Information Retrieval;Relevance (Information Retrieval); *SearchStrategies
IDENTIFIERS *Manual Searches
ABSTRACTA series of five batches of twenty searches were
carried out in a three year span of the INSPEC data base by the 31Company's Information Retrieval Service..The same queries wereprocessed manually in the corresponding volumes of Science Abstractsand comparison of the output of each search method exposed reasonsfor the failures. Emphasis is laid on the fact that a query comprisesother factors in addition to its subject matter and means ofsatisfying these criteria automatically are discussed. The moreflexible nature of the manual searches is examined in detail with aview to improving the flexibility of the computer searches..(Author/SJ)
INSPEC
Report No. R73/13ISBN No. 852964153
U S DEPARTMENT LIFH:ALTH.EDIKATION &WELFARiNATIONAL INSTITUTE OF
EDUCATIONTHIS DOCUe.,ENT lif:f h RE PRODUCED EXACT, v AS RF 17E0 f iv-dmTHE PE,ZSON OR DR,',,NIZ.,,T,ON ORicoNAT,NG 11 POINTS OPIN,DN,
TED DO NOT NE CESS1,11. RE,,RFSENT OF ViciAt NAT iONAI. INSTITUTE OfEDUCATION POSITION OR PO; 'Cy
An experiment in the batch processing of
retrospective 'earches.
H Wilmari
Angela M Hall
February 1973
FILMED FROM BEST AVAILABLE COPY
The Institution of Electrical EngineersSavoy Place, London WC2R OBL.
"C 1972. The Institution of Electrical Engineers. All rightsreserved. No part of this publication may be reproduced, storedin a retrieval system, or transmitted in any form or by any means,electronic, mechanical, photocopying, recording or otherwisewithout the prior permission of the Institution of Electrical Engineers.
SUMMARY
A series of five batches of twenty searches were carriedout in a three year span of the INSPEC data base by the3i's Co Information Retrieval Service. The same queries wereprocessed manually in the corresponding volumes of ScienceAbstracts and comparison of the output of each search methodexposed reasons for the failures. Emphasis is laid on thefact that a query comprises other factors in addition to itssubject matter and means of satisfying these criteria automaticallyare discussed. The more flexible nature of the manual searches isexamined in detail with a view to improving the flexibilityof the computer searches.
ACKNOWLEDGEMENT
The work reported was supported by a grant from the Officefor Scientific and Technical Information of the Departmentof Education and Science.
CONTENTS
1.
2.
2.1
2.2
INTRODUCTION
PROCEDURE
3i Science Information Center
Questions and Profiles
Page
1
3
3
3
2.3 Manual Searches in the Printed Subject Indexes It
2.4 Review of Items Retrieved 4
3. RETRIEVAL PERFORMANCE OF THE COMPUTER SEARCHES 5
3.1 Failure to retrieve 5
3.11. The Number of Profile Fields 'ANDED' 5
3.12 Exhaustivity Within Profile Fields 6
3.13 Bipartite Queries 7
3.14 Comparative Terms and Value Ranges 7
3.2 Retrieval of Inaccurate Items 8
3.21 Term Semantics and Inter-term Relationships 8
3.22 Exhaustivity of Search Fields 9
4. SELECTION OF SUITABLE ITEMS FROM COMPUTER 10OUTPUT OR MANUAL SEARCH
5. ADVANTAGES OF INTERACTION IN A MANUAL SEARCH 11
6. IMPROVING THE PERFORMANCE OF THE COMPUTER. SEARCH 12
7. COSTS 15
8. CONCLUSIONS 16
9. RECOMMENDATIONS FOR FURTHER STUDY 18
APPENDICES
1 A profile coding sheet
2 An example of the computer search output
3 A manual search record sheet
Retrieval and precision figures
5 Summary of reasons for failures in the computersearch
6 Detailed discussion of the exhaustivity of profilefields
8
9
Summary of the inaccuracies of items retrieved
Criteria for the selection of papers suitablefor the enquirer
Suggested criteria for selecting queries best suitedto batch processing and manual searching
10 Useful query information for the profile compiler
11 Guidelines for tape use and profile compilation
1. INTRODUCTION
Over the past few years in which INSPEC has been operatinga computer-based SDI service, a considerable body of experience ,
has been accumulated concerning the processing of the current-awareness query. Less is known, however of the problems andeconomics of searching a data base retrospectively.
The 3i Co, through its Information Retrieval Service, providesfor the batch processing of profiles against the cumulatedfiles of data bases. The addi-zion of the INSPEC file to thedata bases which they currently hold has provided INSPEC withan opportunity to assess, without large capital outlay, thepotential of batch-mode processing of retrospective searches.Consequently, an arrangement was made for INSPEC to submit profilesfor processing on a regular basis over a period of five months.In this tray INSPEC was able to gain experience in the selection ofqueries suitable for computer searching and in the compilationof profiles.
The performance of the st,arches was studied in some detail andSection 3 of this report discusses, with the aid of illustrations,the problems which were encountered. Many of these problems wereconcerned with the compilation of profiles for searching the free-indexed records particular to the INSPEC data base.
In one sense, INSPEC as a data base producer stepped into the shoesof the customers and faced the problems which they experienced whensearching the data base. Although INSPEC has a greater familiaritywith the content of the data base and indexing procedures, it hadno control Over the method of processing the data. Hence, it waspossible to see how' the performance of the searches fell short ofthe potential of the data base because the processors were notfully aware of the structure and content of records.
Of particular interest to the investigation was the comparisonof the computer search with the more conventional search made ina printed subject index. The value assessments made of thecomputer output and the detailed records of .the parallel manualsearches which are discussed in Sections 4 and 5 highlight anumber of the differences between the two.
As a result of the findings of the retrieval analysis and comparisonwith the manual search, a number of ways in which the performanceof the computer search might be enhanced can be suggested. These__suggestions are drawn together in Section 6 and are summarisedin Appendices 8 - 11. It is hoped that they will provide usefulnotes for the guidance of the data base processors and thosewho compile search profiles.
-1-
The costs of the searches although they are unique to the testsituation and not representative of an operational service,are presented in Section 7. The report is then completed bythe conclusions and recommendations of Sections 8 and 9.
2. PROCEDURE
2.1 3i Science Information Center
The investigation was made possible by an arrangementwith the 3i Co. for. INSPEC to provide a series ofenquiries which they would process against the INSPECdata base through their Information Retrieval Service.This service offers to the customer the facility tosearch for relevant references in a choice of thecumulated files of the major scientific and t,echnicaldata bases. The searches are carried out in-batch-modeand process serial files of the document records.
The search profiles may be formulated either as Booleansearch statements or by the use of term weighting, andthe common facilities of term truncation and Searchfield specification are available. The Booleai logicstatements which may be used are of a restrictive nature.Each profile is partitioned into fields such:that termswithin a field are related by the eperaters'ORI and thefields themselves are related by the operator 'AND'.This format is illustrated in the profile coding sheetshown in Appendix 1.
The profile is matched against selected fields of thedata base. In this investigation these contained fre-indexing phrases, subject index headings, classificationcodes and treatment codes. These fields are displayedin the output to show the occurence of term matches.(Appendix 2)
2.2 Questions and Profiles
INSPEC has within its own organisation a library whichserves an extensive community of electrical engineers.The library therefore is continually receiving technicalenquiries and these provided a sustained source from whicha selection could be made for the investigation.
The enquiries, which were recorded by the library informationstaff, were passed to one of a group of INSPEC staff fox'the compilation of a profile. The staff who compiled theprofiles had a wide knowledge of the content of the database but with one exception had no experience of profilecompilation and so a brief explanation and practicalintroduction to profile compilation were given at theoutset of the investigation.
The profiles were sent in five monthly batches of 20'queries to the 3i Co., where they were key-punched andrun against the INSPEC data base for 1969 - 1971.
When each batch of search output was received from the3i Co, an assessment was made (by the information staff)of the usefulness of each reference in response to thequery. This then provided a measure of the effectivenessof the searches.
2.3 Manual Searches in the Printed Subject Indexes
Concurrently with these computer searches, a manualsearch was carried out by the library informationstaff by their usual methods. The results of thesearches, which were supplied to the enquirer as thenormal service,provided a yardstick by which theperformance of the computer searches could be assessed.
For 22 quc.fies a detailed record was kept of the stepsand decisions taken in the search of the printed indexesin the corresponding volumes of the Ab9tracts Journals.A more exhaustive analysis could then be made of thereasons for retrieval failures and the ways in which themore flexible nature of the manual search is revealed.
2.4 Review of items retrieved
Whether a search is carried out in a printed subjectindex or by a computer search the output is usuallysubject to review usually by recourse to the abstract.The information officer may impose additional constraintsin an effort to tailor the output for the enquirer's use,and a number of ways in which this may be accomplishedcan be suggested. For example, a paper may be rejectedbecause it is in an obscure language or the treatment istoo erudite for the enquirer's needs or comprehension.Following the next section, in which the performance ofthe computer search is discussed, attention is given tothe bases on which this review is carried out, and oncedetermined some may suggest alterations to the computersearch to improve its performance.
'3. RETRIEVAL PERFORMANCE OF THE COMPUTER SEARCHES
3.1 Failure to retrieve
The overall performance of the computer search wasdisappointing, because a considerable number ofsearches were completely unsuccessful. Twenty-three of the 100 searches retrieved no references anda further sixteen retrieved only unsuitable items. Thisfailure rate was particularly high in comparison withthe manual searches, since in the final 22 searches, 8computer searches retrieved no references whilst only 1
of the manual searches retrieved no references. Sim4larly14 of the computer searches retrieved no suitable it,ms,whilst only 4 of the manual searches provided nothingsuitable.
With hindsight, some of the causes of the failures ofthe computer search can be isolated. Comparisons with theparallel manual searches in particular, serve to exposesome areas of failure and suggest means by which thesemay be overcome.
3.11 The Number of Profile Fields 'ANDED'
The results of the investigation emphasise that, thecomputer search is, in general, no better able than themanual search to satisfy the query which is expressed bya combination of a large number of concepts. The mostcommonly occuring reason for the failure of profiles toretrieve any references was the requirement of a match infour or more profile fields. This was instrumental in thecomplete failure of at least 16 of the 100 searches, andpartial failure of 41 searches. It is possible thatthe data base contained no suitable items, but the factthat the majority of these searches were satisfied by themanual search indicates that the exhaustivity of indexing,although higher than that of the printed index entries, isnot high enough to satisfy profiles of this depth.
The exhaustive profiles occur most commonly because theinexperienced profile compilers expressed every non-trivialterm of the query in the profile. This pressure to satisfyevery concept in the query is weaker in a.manual search,because the searcher knows from experience that the printedindex entries are not sufficiently exhaustive for thisapproach, and he accepts a lower level of match. Thisrelaxation of requirements does not apparently lead to theretrieval of much noise. For example, 5 of the final 22profiles comprised four or more profile fields. Onlyone of these profiles, matched automatically, retrieved
a document and th:.3 was not relevant, but all thequeries were satisfied by the manual search to aprecision of 62%, i.e. 62% of the index entries selectedwere judged to be suitable when their abstracts wereconsulted.
The most successful computer searches were those whichdemanded a single concept, e.g. 'auto-closing circuit-breakers'. Queries of this nature may normally beexpressed in two or three search fields. This was thesituation with 27 of the 35 successful searches. It isalso interesting to note that eight of these comprisedone of the printed subject index headings qualified by asingle adjective and would, therefore, have been quicklysatisfied in a manual search.
The frequency of occurrence of the many-termed query is oneof the characteristics which distinguishes the queries inthe present investigation from the current-awarenessprofile. Although a current-awareness profits may includea large number of concepts these are often alternativesand not required simultaneously. It is suggested by theresults that it is not natural for a profile compilerto write a profile which is less specific than the query,but that he must be encouraged to do so in the inter :stsof retrieval. A number of considerations for minimisingthe number of neccessary search fields are listed inAppendix 6.
3.12 Exhaustivity Within Profile Fields
Whilst increasing the number of 'ANDED' fields in aprofile reduces the probability of retrieval, increasingthe number of alternative terms within a field increasesthe probability of success. These alternative termswould commonly be synonyms, or related terms. For example,a profile of a query about 'metal foils' has improvedprobability of success if a list of named metals is addedto the field 'metal', e.g.:
Field 1 Field 2
Metal FoilAluminium FilmSilverGoldCopperIronNickel
This example is straight-forward but there are someterms for which synonyms or specific examples are lctsseasily defined. For instance, it is difficult for theprofile compildr to construct a comprehensive list of the
-6-
'Electric hand tools' which cause interference. Theprofile compiler must, however, attempt to do so becausethe absence of control of-document indexing at inputpermits the listing of any tool by name.
The investigation supplied examples of other unsatisfactorysearch terms, e.g. 'equipment' and 'device'. The indexercustomarily describes a piece of equipment by name and notin these general terms. An even greater problem is the useof subjective terms such as 'development' and 'advances'in a profile, e.g. 'developments in oscilloscopes'. Thereis little possibility of enhancing a profile with theseterms. If they are included in the profile then the itemsretrieved will be of the review type and particulardevelopments will not be retrieved since they are unlikelyto contain these terms. If on the other hand, the'development' apsect of the query is omitted and only'oscilloscope' used in the profile the retrieval will belarge and will contain a lot of unsuitable items. It isprobable that an enquirer who couches his query in termssuch as 'advances' or 'development' will be best pleasedwith a small number of review papers. If, however, herequires a comprehensive search, he would prefer the largeand crudely selected output in order that he might makethe final subjective judgements himself. It is veryimportant to the profile compiler that he should knowwhat purpose the enquirer will use the reference and havean approximate idea of how many papers he is able to read.
This information about the enquirer's requirements ismore important for the retrospect:ive search than for thecurrent-awareness service for which the purpose is normallya watching brief with a practical limit to the outputsize dependent on the effort the user is willing tospend in following up the items. This effort generallyremains constant from week to week.
3.13 Bipartite Queries
Another type of query which made profile compilationdifficult was that in which the enquirer asked initiallyfor papers with a braod subject coverage and then expressedinterest in a few specific examples, e.g. 'Doppler effect -acoustical, optical, radar - especially re-unscramblingthe return signal from various types of interference'.
This dual appraoch basically constitutes two profiles, andagain if the enquirer's purpose is known, a more usefulprofile can be compiled.
3.111 Comparative Terms and Value Ranges
Eight of the queries included terms such as 'high frequency'and 'small scale'. These are terms which are used in theindexing but whose use in a profile is more difficult.
The indexer will not always consider this aspect ofa paper sufficiently important for inclusion or he mayexpress it as a numerical range e.g. 3,000 - 30,000ICJIz.The search system is not equipped to compare themagnitude of numbers. Unless the range in the profileis exactly the same and is expressed in identical unitsno match is possible. Even should numerical valuesbe converted to the more general terms e.g. 'high frequency'this also can be confusing since high frequency meanssomething different to the audio specialist and radioengineer.
3.2 Retrieval of Inaccurate Items
The deficiencies of the computer search lie not only infailure to retrieve references, but also in the retrievalof references which satisfy the search profile but byno means satisfy the question stated, or the enquirer'sneed. Precision calculations for the computer showan average value between 40 and 60% depending on themethod of calculation.
The reasons for the retrieval of inaccurate items fallbroadly into the two categories (i) indexing failures and(ii) profile failures. Those which occurred in thisstudy are summarised in Appendix 7 and some are discussedin the following paragraphs.
3.2.1 Term Semantics and Inter-term Relationships
The problems associated with the meaning of terms and therelationship between them are complex and the distinctionbetween the two is not well-defined. The former areparticularly prominent in a system in which the indexinglanguage is minimally controlled. Firstly, the meaningof a term can be highly dependent on the context of thedocument in which it is used. For example, the termtlinethas a different meaning in the context of 'power lines'and'assembly lines'. In many cases the discrepancy isovercome by searching in only a limited part of the database, but this is not a complete solution. In the samecontext authors or indexers will use one term in differentways, e.g. in a paper about train control the term 'remote'implies centralisation to some authors, but not to all.
Secondly, the consequence of searching the free-indexingfields as a string of unrelated terms instead of a seriesof distinct phrases is marked. The relationship betweenthe phrases is not expressed and they frequently referto completely different parts of the document described.Two terms each retrieved from different unrelated phraseslead to inaccuracies. A query concerning 'current' probesin the measurement of current, with the profile:
Field 1 Field 2
probe current
retrieved a document decribed by 'Current controlledsources; Power supply circuit; Range switching;Temperature probes; Bridge circuits.'
In more extreme cases one indexing phrase may describea very minor part of the document. Terms such as 'control'and 'circuit' are particularly subject 4o inaccuracies ofthis type. This was a common occurancc, producing somenoise in 23 of the searches made. It may sometimes beprevented by the co-ordination of search terms in a singleprofile field. For instance, in the example given above'current problel would constitute a sir7le field and must,therefore, retrieve only those records .n wnich the twowords occur cons.cutively. In practice the output of theprofiles in which this device was used eras very small.
The preceding failures concerned the absence of anexpressed -elotionship between terms, but the situationalso arises in which the relationship between terms in asingle phrase is expressed in the document indexing:usially by a preposition such as 'in or 'on'. The Booleano:-erators cannot distinguiSh this difference. Hence,in a search for papers about 'patient care after shock' anumber of papers were retrieved which discussed 'shockoccuring during patient care'. These inaccuraciesoccured less often in the investigation, but were notedon six searches. It is, however, sometimes difficultto distinguish between two types of failure.
3.2.2.Exhaustivity of Search Fields
The retrieval of items which do not accurately match thequery also occurs when concepts of the query are omittedfrom the profile. This happens if a concept is difficultto express, for example, 'applications' 'developments'and 'high frequency'; all concepts which have already beendiscussed.
More commonly, any omission resulted from misunderstandingbetween the profile compiler and the information officer.This was a consequence of the artifical test situation inwhich in view of the large number of questions involved, itwas necessary for the information officer to delegate thecompilation of the profiles. This would not be the case ifa continuing service was offered, but it serves toillustrate how ci.:;ily an important detail of a query may belost at the expen.,e of the search results. In Appendix 10therefore, a check list of necessary and useful enquiryinformation has been compiled.
Lt. SELECTION OF SUITABLE ITEMS FROM THE COMPUTER OUTPUT
The number of references retrieved in a computer searchcan vary from a single item to some hundreds dependingon the breadth of the subject matter. In the latter casethe search might be thought to be 'too successful' and mustbe screened to yield a practical number of references, Thelists of references sent to the enquirer as a result of themanual searches indicate that 6 to 12 is, with a few exceptions,the number generally accepted as practical. The rigour with whichthe screening is carried out will depend on the size of theoutput, and where the output is very large only a small portionwould be reviewed. In the following paragraphs, however, thebasic crLteria by which this was achieved for both the manualand computer searches are detailed.
4.1 Selection Criteria
The language of the paper is the most common criteriaused for screening, and where a number of referencesin English are available they are often selected inpreference to foreign language ones.
The papers selected may represent a range of the moregeneral and specific aspects of the subject matter and areview paper may be included together with avariety of applications, techniques or devices. Thismany-sided approach is necessary if, as frequently happens,neither the enquirer's purpose or level of familiarity withthe subject matter is known. If the enquirer's familiaritywith the subject is known, then specialised papers, perhapsor papers with a particular slant will be rejected. Othercriteria used slightly less frequently were the age ofrecords, the country'or organisation of origin, the standingof journals in which they are published or the type ofpublication, e.g. report, thesis, etc. All these affecthow readily the full text may be obtained, although this isnot the only reason for their use.
It is, incidentally, these criteria of selection whichrepresent another distinction between retrospective searchand the current-awareness service. Although they areapplied to some SDI profiles at the user's request, thenormal service provides a less restricted output.
These criteria are listed in Appendix 8 and also constitutemuch of Appendix 10 which acts as an aide memoire for.ensuring that the profile compiler has the maximum possibledescription of the enquirer's requirements. In Section 6,which suggests some means of enhancing the computer search,the application of this information is discussed.
5. ADVANTAGES OF THE INTERACTION OF MANUAL SEARCHES
The previous section discussed the action taken with largescale computer output. In this section the situation atthe other extreme is considered. The output might be verysmall, even non-exis,ent,and cannot be improved except byrewriting the profile and repeating the search. In a manualsearch the adjustment of profile or requirements is carriedout continuously throughout the search in the light ofresults. It is instructive therefore to study the results ofthis interaction in anticipation that some elements of theinteraction nay be simulated in an automatic search.
The selection of items in a manual search is not restrictedto the entries which contain all the significant words ofthe query. The user will select entries in which a relevantpaper is expressed irrespective of the terms used.
It has already been noted that the free-index term assignmentis insufficiently exhaustive to satisfy the profile withmany concepts. The headings and modifier lines in theprintea subject indexes are less exhaustive and so have nogreater probability of.satisfying these queries. In orderto avoid a non-productive search the user relaxes theconditions and this is commonly achieved by selectingentries which match only some of the concepts. This approachwas adopted in eight of the final 22 searches with littleresultant noise, since a precision of 62% was recorded.
An alternative approach adopted in four searches was theselection of index entries describing associated techniquesor subjects)or documents indexed by attributes which differ fromthose of the query. They are selected in anticipation thatthe paper will prove to be relevant when more details is available.
It might have been expected that the user would select indexentries which include terms more specific than those of thequery, but this was an uncommon occurance, which occured onlythree times in the whole investigation.
It can be concluded that the variation in the manually selectedentries from the query statement generally lies in the reductionof the number of concepts sought, and we can consider theapplication of this result to the improvement of the computersearches.
6. IMPROVING THE PERFORMANCE OF THE COMPUTER SEARCII
The presentation of results and the discussion of thefailures of the computer search have highlighted a numberof aspects of the performance which might be improved.For example, by performing some of the operations in theoutput review procedure or in accommodating the query withmany concepts. In the following paragraphs some suggestionsare made for the implementation of such improvements.
6.1 Variation in profile field exhaustivity
Although the 3i's search system includes a facility forthe weighting of search terms, this was not utilised inthis investigation. Each profiler compiled only fourprofiles each month and since the majority of them had noprevious experience, it was not reasonable to expectthat they would master two alternative search techniques.This facility does, however, permit a more suitablesearch strategy for those queries which comprised a largenumber of concepts. Appropriate weights are assignedto each term or field and the document output is rankedby the sum of the weights of the matching terms-. Theitems at the top of the list then satisfy the querycompletely and those lower down satisfy it only in part.If then no items satisfy the query, some less exact oneswill still be retrieved. This result can be achievedin its simplest form oy assigning each field a weight of1.
The more complex problems encountered, of the overlap of conceptsbetween fields, can also be overcome by the use of weightingin preference tp the Boolean formulation. Although itshould be pointed out that the problems were due to therather restricted Boolean formulation that was allowed andwould not have arisen had a more flexible version beenavailable.
Throughout this study although the manual searches wereconducted on the basis that the enquirer was given onlyas many references as he can reasonably cope with, theprofiles were not compiled with this in mind. The Users'requirements were not always known to the profiler and althoughthe experienced profiler can, with a knowledge of the database, vary the profile terms and exhaustivity to produce adesired output size, their success in this is severelyLimited by the 'all or nothing' requirement Of the Boolean-based formulation permitted. Again, in this search systemthe weighting of search terms is one answer to this problemand the output size may be limited either to a specifiednumber of references or by a predetermined threshold of matbhingweights.
6.2 Language of reference
For the majority of queries, it is essential that thepapers retrieved should be written in a language whichthe enquirer can read or for which he can quicklyobtain a translation. This fact has already beenrecognised by INSPEC and language information has beenincluded in the records on the data base. When anattempt was made to use this information in theinvestigation, a particular difficulty was exposed. Onlyforeign languages are entered in the record. If thepaper is in English the record field denoting languageis generally left blank. Should the enquirer requestonly English Language papers, as is often the case, itbecomes necessary to search for an empty or non-existentfield. The search system used did not permit this. Thedata base processors were obviously not aware of theproblem, which would be simply solved by inserting'English' in the blank language fields before the data isprocessed.
6.3 Treatment Codes
INSPEC had also anticipated that it would be helpful tobe able to retrieve only those papers which treat thesubject matter in a particular way, for example a reviewpaper, theoretical treatment or developments - arequirement which has already been discussed in somedetail.
Forthis purpose 'treatment codes' are included in therecords where appropriate. This policy had not beenadopted when the early records of the archive wereindexed. Thus, if a treatment code is necessary tosatisfy the logic of a profile, none of the early recordswill be retrieved, even though they may treat the subjectin the required manner. This problem is overcome onlyif a match on the treatment code is not essential to theretrieval but adds to the final output rank of thedocuments which include it.
The 'treatment' factor also has a more complicated aspect. Thevariety of papers which the manual research provides, includinggeneral and specialised papers and a selection of applicationsor techniques, is difficult to simulate. Although a cleardefinition from the enquirer can greatly improve matters,it does not provide the complete solution.
6. Books
The book as a source of general articles was frequentlyused in the manual search. They are quickly selected fromthe library shelves and are more likely to cover the well-established subject matter than are journal articles. Theyalso often provide bibliographies. The computer, search
-13-
does not easily manipulate the book entries. Theindexing is very general and is rarely specific evento the extent of chapter headings. A section or chapterin a book, although relevant cannot be retrieved by aprofile designed for the retrieval of journal papersand reports. It is questionable whether it is of valueto include books in the searchable data base when adifferent search strategy is required.
6.5 Other search fields
The information officers' assessments stressed a numberof other details which are sometimes of interest in aretrospective search. Amongst these are included thedate of publications, the country or organisation of originant[ the source journal or type of literature. All thisinformation is available on the'INSPEC tapes and couldwith .7ome attention to suitable search strategies, beuseful search fields.
6.6 Summary
It can be seen that the majority of features to which themanual searces owe their flexibility and selectivity canbe simulated in a computer search by use of more of theindexing fields available on the INSPEC files and by useof more flexible search formulations. A more flexibleBooLean logic or weighted search system would solve manyof the problems presented by profile compilation in thisinvestigation, but some of the errors will also be overcomewhen the INSPEC thesaurus, displaying the relationshipbetween the terms is available to the user.
-14-
7. COSTS
The average computer processing cost of each automaticsearch was £23. A further £2 might be added in respectof the profile compiler's effort but this is only a smallproportion of the cost. The cost of the computer searchis of an order of magnitude higher than the manual search.
The manual searches it the printed index took an average 45minutes and would therefore be unlikely to represent a cost ofmore than £2. The cost of assessing abstracts is ignoredfor the comparison since it is common to both methods ofsearch).
The effectiveness of the computer search must be high tojustify the cost, but in this experiment the completefailure rate by this method was eight times that of themanual search. Although means may be suggested for reducingthis failure rate, the noise would be also increased andadditional manual selection effort would be required.
8. CONCLUSIONS
The overall performance of the computer search was not goodwhen compared with the search carried out manually. Therewas a high incidence of completely unsuccessful searches anda number which produced very large outputs, of the order oftwo or three hundred references.-
The cost of the computer search was of an order of magnitudegreater than the manual search in this instance and it is knownthat the price charged was significantly lower than would havebeen charged for an operational service.
Two or three weeks elapsed between the dispatch of the queriesand receipt of the output and although this was due to thedistance involved it is unlikely that even an inhouse operationwould achieve a turn round on batch processed searches of lessthan a day. Some difficulty was experienced in the later stagesof the investigation in collecting sufficient suitable queriesand it could not be expected that enough queries would berecorded in a single day or possibly even a week for batchprocessing to be eeonomic.
The investigation showed that there is more to a retrospectivequery than the subject matter. The enquirer does not normallystate the use to which he intends to put the information, thelanguages he is able to read, or the type of literature to whichhe has ready access, and in written communications he onlyinfrequently indicates the level of treatment he requires inthe documents retrieved or any particular bias. Information ofthis type is extremely important to the user of the printedindex or profile compiler, but is often not consciouslydetermined by the information officer taking the query. Thepresence of two intermediaries between the enquirer and hisprofile therefore, led to the loss of some of this information,but in an operational service this situation would not arise.
The searches most successfully completed by the computer searchare those whose subject matter comprises a single concept andnot those combining a series of concepts as might have beenanticipated. The indexing in the free-indexing field of thedata base was,not sufficiently exhaustive to satisfy the mostexhaustive profiles. In order to increase the retrieval it isnecessary to relax the conditions and select items indexed byonly some of the profile terms. This is an approach which ismost commonly followed in a manual search to increase theretrieval where printed index entries are insufficientlyexhaustive. Unfortunately, an.increase in the number of itemsretrieved almost inevitably means an increase in noise, andgreater manual selection effort is required to screen the -Jutputto provide the user with the most-useful list of references.
Ultimately, a reduction of profile length produces a listof references or screening which is equal in length tothe entries under a printed (subject) index heading-; Theunsatisfactory performance of the long term profiles servedto emphasise that care must be taken when a profile iswritten, that query concepts are not repeated in more thanone field.
The failures of the computer search lie mainly in two areas.Firstly, the inability to retrieve references in responseto long profiles and, secondly, in the retrieval of inaccurateitems as a result of the lack of expressed relationshipsbetween terms.
The logical formulations available in the 311s system displaysnot only those term-relationship problems typical of Booleanlogic, but also problems resulting from the inflexible format.This can be overcome by sophisticated profile compilation orterm weighting but neither procedure would be expected of theinexperienced profiler.
Analysis of the manual searches made showed that few intellectualdecisions are applied in an effort to increase the number ofreferences retrieved, but much effort and many decisions aremade to select the most suitable references from the longerlists of items retrieved. This fact is encouraging for theimprovement of the computer search as methods are availablefor simulating the relaxation of profile requirements and thescreening of output.
These operated in batch-mode, could enhance the performanceconsiderably but they are still restricted by their one-offnature. The advantages gained by user system interaction andits consequent profile modification which are so importantin the manual search cannot be achieved in batch processedsearches but are to be sought in on-line searching.
9. RECOMMENDATIONS FOR FURTHER STUDY
This investigation has shown that the main areas in whichfurther studies are necessary are in methods of satisfyingthe exhaustive query and in increasing the selective powersof the computer sear h.
The former will be achieved by investigation of the rankingof output and search logics which do not have the 'all ornothing' limitation of the Boolean statement. The optimumuse of the classification codes would also be closely linkedwith these studies. The classification like thefree-indexingis unique to the INSPEC data base and so although some studieshave ;,een carried out elsewhere concerning the best approachesto retrospective searching, the results obtained are notnecessarily also applicable to the particular form of indexingused by INSPEC. For example, the optimum use of the discreteindexing phrase is not clearly understood.
The study of the optimum use of the treatment and languagecodes on the INSPEC data base are not subjects for extensiveprojects but deserve some attention to detail. Satisfyinga query for a range of treatments is undoubtedly a problem.Searching numerical ranges and comparative terms again needsome investigation.
The usefulness of searching for books and papers by the sameprofile was brought into doubt by this investigation andwould bear further consideration.
-18-
PR
OF
ILE
DE
SC
IIOP
T10
14.C
O-c
HIE
NT
S.
LEE
RD
AT
E.7
.1P
AG
E:
1_
OF:
EE
TR
OSP
C.,T
IVE
SE
AR
CH
EX
PGR
IME
hyr
ri r
siT
Or¢
AT
TA
CH
CO
OIN
G F
OR
MS
FO
R A
LL C
OR
RE
CT
ION
S -
IND
ICA
TE
DA
TE
S O
F C
OR
RE
CT
ION
S
CO
.N
O.
jP
RO
F.
NO
.
x C.I
ti
DA
TA
,AR
TE
Apo
e,pA
ST
IEX
DA
TA
BA
SE
IAC
CIG
1a.
s g
PR
OF
UN
ITS
TH
AE
ss,
CO
Mm
EN
TS
01/
FX
.99 jA
. ,.1
.5* 7
iE
Em
igill
allg
igel
imm
ula
1211
1.12
1321
.11"
1=1:
1711
1:21
=12
1==
r211
-1"1
111-
1211
=T Y P E
;IIt
Iris
eirs
auss
imm
EE
3E-
imis
imcm
iztr
a=6.
1.14
.us
issi
onm
umar
ea1/
P A a A M
L G i C
s G N N
1
GT
. 0 0 E
i
.
TE
RM
VA
LUE
(T
O C
ON
TIN
UE
04
NE
XT
CA
RD
RE
PLA
CE
:`,
BY
./1
i 4 6t )'2
3'9
'_4
0 11
2.3
7.
aO O
DO
45
67
. 8 '
90
OR
M5
0128
9'0
24
56
7P
.91
01
230
5,6
T18
.9 ;0
1, 7
i3. 5
.6 :2
'0.9
'0am
mr1
138
9O
man
,,r
.7
019!
0
la-
,111
1
-_ s.:- 4.
1i
Hili
mI
1111
1111
1111
EM
LIM
IIIM
MI
:11
1
1111
MillE
N IM
I
,:
,.
.,
,,
,
..__
, , ,...
.
i.c,
NE
1111
1111
1111
1
MIM
I11
1111
=11
1
,1
1
!
,.
1I
:
!1
t1
..
1 ,! , .
i
1111
1111
11
II
iIf
s s
/i
T..i
t
dt.
,
I? 1
IL
izI IL
..
!
)'
111
IIII
till
al
.0'.
..za
'ik
"C-X
:1'
....
_.!-
,,
nli
:& '-
:01
/ a
1
nu 1111
1111
1111
111
1111
11E
11
II
'I
1
1
.
t
. ..
.
I1
I
;I
'I
i,
1I
I
MII
IIII
MM
IIIM
IIII
IIII
IIII
MI
MII
IIII
IIIM
S111
1111
1111
1111
1111
11m
umMill
iH_
S I
f,
/it.
'
11p.
- L
i'!
nim
in1
..
:
r
.:
'
mom
min
1111
11M
IIM
UIM
. .,
s,,ul
9 111
11
111111 II
7,.;,
r1.
I
Mill
I 1
1111
1i
Ia 1' %-
,--
I
n / .
I1--
,,t--
1....
-cti,
1
OH
M11
111
:
. .
1111
1111
1111
1111
1111
111
1111
1111
1111
1111
1111
ill
1111
1111
1111
1111
1111
1111
'
r.r.
.
INN
i.s
1
1111
s., i
m!
4-74
,.
.
1111
1111
1111
1111
11.1
1111
1111
NE
M..
i!
;:
:.
,s 165
, ;
1111
,.
.
/4 M
' PI
4*:
IMM
IUM
1111
1111
1111
1111
1111
11=
11!
'.
i!
.
;;
;.
i1
!:
i1
:
1111
1111
1111
11%
ai
1Ill
1_
ii
5 '
117 le 19 --,
s I
i I-
1t
=M
ill IIIM
ULU
MI
NE
M1
111
I
.i' w
i!I
,
1111
1111
111
1111
1101
11.
i
SI
_III_ ,.
1 !
,---
-,1
._
CO
urA
Ny/
0..F
C.P
4/.5
.1,O
N,Ic
rTS
RS
CIE
NC
E IN
CO
RP
OR
AT
ED
2711
PJa
cx-,
a,H
. Pa,
'51C
13S
CIE
NC
E !i
.-O
RM
AT
ION
CE
NT
ER
PR
OF
ILE
C1
'11 0 H t- tr1 C) 0 CI H z
APPENDIX 2
COMPUTER SEARCH OUTPUT
1:0CW1:T T:1P i;r iyi rt,',1.
2CCC.56 999/06 Oen 00003 UTf: 13Jui.72 2
TITL7t The 1,10.nrnent. oR an Iqrficid tiy libt/0;77-10; '45.charcini
Or.C, -h,n[ziands
(J ) Pcs, Tee1n,o1 (GN VOL 2, NO°3 036-.:!"
CLA:ISIFICATT0N: Pi1630
Fi:Lnr!! l'AlOrtJpr (110M n1rfl.ftld tn;<11.1R,(if.)n7J-c!'lt!; Pn( P'or.Pirr.1.)(1
Iccqh:ronf;n) /1;yi.r.contv(;) (111 IT:quiPeentap runwaY an6 trai;),1Y ii111)11.n;itinoD.c:vc)i.0/
!;(3'a,C ArAG75 PEF0-,:flOo/TRFATVEnT: 870398)6
,APST'OACT: ilevelonw:hts'ir opcilmtiona1 procc!6uret.
cocurNT No. ppoyITE Tim ACCUH WGT DATA 1Wt! cAPT: 2999/C76 000.1 00003 IEg 13,1002 0;!' 2
zkrimw oth(;r thinrr.is, an abfaitY to ta.x.1 tflcf, Yom runNay 4,0 the unloaen..7 !Irea 111.1r- thc:
f!c:1'101(,n:; of. h;.J a (I fl(
tftYiwr;Y :0j,:htn v:c(!t stri.vw,'nt0 Tec:nno). C f:rue-.1j.Ar:;:ti0n, The. v;aper ccvoArt
;0)01 .c:LyAinr. t7.:1:%.(1 on 11,.! of tun).:0!;(.,rnInr;i.t:Irw a 1o'4, very cc;:iract c2nidco PY(,):;)
ttillTh fl J hnelrrotilf: tot or vivc::no tc% (i.te!,::ilat-Infactory perf.ornanee
1.0
0
L
IEE
PROJECT AB RETRIPNAL CFNTR2
MANUAL sEAncw BPcnrwl. 1^7T-C`q'rTn-NT ,Tel 5?"1
Stn. t- of Encluba....41. Original letter etc.
/D/L on
2. Statomunt of query
Pap 6,./-5 0 /1.
10-1-5(-c /o. Ai r..o
- i-ra p o n.3. Enquirer
and /of rrto Ps
cC tes La is Arz,
n eas Ltreirt C4"1-/ /-eS /1-1
rrvac1-1 I nes arui /7nrtslorol(Aeck c.. I:, te.a.1; c c
4. Enquirerta background
of
50 Rontrietif)n3 oo language, Bizo of output, troatmont., (Irt;bibli:Igraphic form. (Thor ° reatrictionn may alY2licrA
inforr,kation officer in hiu experience on beh:O.T of ther-01030 are not rentrietiona or rolwzations Lupo:ledduring tho aearch. GC° II. .
6. Rallon for ao/ootion of query for toot
suiAt.a(c
Query rocolved by k/31):.%te.
DK:arch conduct:DO. by Av.;
c000tion No.Procnrhlre
Volu:::on of Scionco Abotracto noarchos
fi7.1 r, c / a,ta ec,krort Ls iro-
9. Indox hoadinR3 urod.
krx.6 a4-13/6.. Me-0-9--Ltecr-1. e-rk,
/ M
10. Li3t of abstract numbers oolocted from modifior lino ti. Itom3aro ansi(mod to ono of tho following catogorim (A r4 accurui;nS = Gultablo 1.o. aondablo W o Not).
i) S (and NA) :44Accopt All itomo in cator::orioo
ii A cnd S Accept (i) (lii) arfa log,;cd
ill. A (and NS) LI A, Rojoct in colmtn I (1))5e.)iv NA Reject
Loe
Hoadln Hoc cling
arl-de- %>-i u-i.
?c,/3
Y3
3`t 36,5-
.38 /3/6 ?Y3
-808/.1.38 '
.5:5a, e.
/71-S
S
/9 /.S
N5ArS
A/
A 1/3
15(S
19 IS
445
PO1,311
Ii/W6
A ea...nun/ an_
V.0
3. 2
Rol;trictionz or relaxations imposed by the searcher,
- /ti," er c) ai-erzis
Ab-str!7-.c t%3 or accepted items from 10 are examined and eachite: is annied to one of the 4 categories on the basigof the infoalion in the abntract in CO]. 11. Give roaoas Q-ABC)
13°ito748 found in Science Abstracts in any other way thandivoctly through the index and how.,
14. it from other sorvices.
15. Detcylls sent to enquirer - copy attached
1. 'Tine for looking at luodifies lines 30 /rum:,
om uter Search
Manual Search
QUESTIONS
No. of
references
retrieved
(a)
No.selected
by inform-
ation
officer (b)
Precision
(a)/(b)%
No. of
references
retrieved
(a)
No.selected
by inform-
ation
officer (b)
Precision
(a)/(b) %
90. Use of computers for the calculation of
ground potential distribution in high
and medium voltage substations with a
ground grid and rods.
91. Capacitance braking for 3-phase a.c.
motors up to about 10 HP
92. Installation of standby generator sets
and connection to feeders and isolation.
93. Impedance.of coils.
94. Partial discharges in dielectrics
95. Fault diagnosis in combinational networks
96. Divided-winding-rotor generation.
97. Design of coils windings etc. for
transformers and motors.
98. Precautions taken in laboratory wiring.
100.Future -uses of electrical heating.
5 0 0 1 0 6
56 0
184
0 0 0 0 0 6
13 0
22
0 0 o/o
0 100 ++
0 100
.23
o/o
12
8 9
14 9
15
10 4
14
10 9
1 6
11 9
14
10 4 8 2 8
13
67
79
100
93
100
100
57
20 89
Computer Search
Manual Search
QUESTIONS
No.of
.
references
retrieved
(a)
No.selected
by inform-
ation
officer (b)
Precision
(a)/(b)%
No.of
references
retrieved
(a)
No.selected
by inform-
ation
officer (b)
Precision
(a)/(b)%
64. Use of thyristors in speed control of
motors in the Apollo Space Programme.
67. Prosthetics for knee and hip amputations
i.e hinge-joints for ex-articulation.
76. Landing lights for airfields - design of
lamps, installation and cabling
78. Continental practice for electricity
meters - designs and use of polyphase
meters.
81. Bibliography on synchronous induction
motors, rotor excited from static sources
82. I-diagrams - the measurement of noise and
interference on digital waveforms by
oscilloscope methods.
83. Loss angle measurment, testing of
insulation etc. of motorsjllachines/
transformers and extrapolation of
results to predict breakdown
84. Technological trends in electronics e.g.
LSI, IC's, computers, communications,
data transmission facsimile and components.
85. Jointing of undersea cables under high
pressure
86. Speed control of fan motors (split phase
capacitor start) using thyristors.
88. Manufacture of very small high speed
(over
10,000 rpm.) d.c. motors running on
4-6voLs
using a wet cell charagable battery.
89. Automatic synchronisation
of generators.
0
10 7 2 , 0
35 4
83 0 9 o 3
0 8 7 0 0 0 o
67 0 5 o 1
o/o
80
100 o
o/o
o o
37 ol0
56
5 ofo
33
1 7
11 9
21 3
15 6 7
18 1
1 4 o
0 4 8 1 5 o 6 3 4 8 6 o
0
56
73 11
48 o
ho
50
56
'*c c
44
tx. 2 t 1--
43
),
ojo
APPENDIX 4
Itetrieval and precision.figures
Questions Computer search retrieval.
No. of No.selectedreferences by inform- Precisionretrieved ation a/b %
(a) officer(b)
1. Medical thermography
2. Small-scale applications forlinear motors
73. Electronic speed control ofgas and steam turbines
11. Centralised train control
38
0
1
38
30
0
79
o/o
1 100
20 53
5. Current probes used inmeasurement of current 73 60 82
6. V.L.F. undersea propagation - +
7. Measurement of iron losses2 1 50using wattmeter and 'ghost coil'
8. F.E.T. oscillator circuits 19 12 63
9. Cable balancing using a compute 24 0 0
10. Antilock brakes 20 18 90
11. Dopplereffect - acoustical,optical and radar - especiallyre-unscrambling the return .-+signal from various types ofinterference.
.12. Insulators for overhead lines andtransformers
13. Quadrature amplitude modulation
14. Vehicle guidance by signalsreceived from underground cables
15. Demagnetizing devices
16. Medical ultrasonics: particularlynarrow beamwidths transducers andtechniques for measurement ofpolar diagrams
17. Design of 3-phase to single-phasestatic balances, especiallyapplicable to single-phasefurnaces
18. Speed-control of polyphaseinduction motors using triacs
19. Protective multiple earthing
3
0
0
10
0 0
0 o/o
20 ++
- +
0 0
10 100
20. Autoclosing circuit-breakers - +
APPENDIX 4 continued
Computer search retrieval
No. ofreferencesretrieved
(a)
No.selectedby inform-
ationofficer(b)
Precisiona/b ();.,
21. Floating point hardware
22. Modular numbers
23. Live-line washing of insulators
24. Computer-controlled traffic lights
25. Engineering technician education
26. Electrical installation in nuclearpower station buildings
27. Lightning protection of buildingsusing the reinforcement inreinforced concrete
28. Eddy-current couplings
29. Defibrillation after electric shock
30. Boxcar integrators
31. Dimming of fluorescent lamps
32. Suppression of interference fromportable electric tools
33. Solid state control of Ward Leonardmachines
30
3
1
8
5
1
0
1
5
1
0
2
24
2
1
7
5
0
0
4
0
0
2
80
67
100
88
100
o/o
O
80O
o/o
100
34. Sodium-filled cables
35. Gasturbine alternators
36. Peak load lopping using dieselstand-by generators
37. Metal foils for electrolyticcapacitors
38. Electric actuators
39. 1Lincampextradio-telephony system
40. Telegraph message switching withelectronic stores
41. Power transformer audio noisereductlon using antiphase noisetechniques
42. Speed control of induction motorsusing thyristors
43. Synthetic testing of circuitbreakers
44. Description of the Decca 'Navigator'System of radio-navi&ation
45. Lead/acid battery maintenance duringperiods of non-use
46. Power system interconnection
47. Plasma arc cutting'of steels
9
14
11
4
7
13
7
4
78
93
64
6o ++100
1 1 100
0 0 o/o
22 21 96
28 27 96
2 2 100
2 1 56
5 ++5 4 80
APPENDIX 4 continued
Computer search retrieval
No. of No.selectedreferences by inform- Precisionretrieved ation a/b
(a) officer(b)
48. Long term reliability of discrete-component electronic semiconductor 5 4 80equip. (cf'bath-tub' characteristicsof vacuum tubes)
49. Load-frequency control in inter-connected power systems
50. Developments in oscilloscopes
51. London's power supply system
52. Anglo-i,rench hvdc link 0
53. Electrical power development inSpain
54. Push-button telephones - their
6
5
3
0
6 100
1 20
2 67
0 0
0 o/o
design and uses especially in datatransmission
55. Lifts in high-rise buildings
56. Ultrasonic atomisers for theproduction of aerosols
57. Surge protection (diverters) forelectrical systems
58. Computer typesetting
59. Conduction in graphite-impregnatedrubbers and plastics
60. Computer programs or software forretrieval from documentation databases
61. Invalid carriages for disabledpersons
62. Electrical parameters of earth soiletc. upper layers of interest reradiowave propogation, especiallymaps of G.B.and Germany
63. Ripple control of power systems anddomestic equipment, lighting etc.
65. Use of electronics in the fishingindustry e.g. data processing andlow light T.V.
66. Electrical services in hospitals
68. Substation earthing
69. Standardisation in video recording
70. General papers on stepper motors
9
6
3
2
20
18
2
0
18
11
11
0
8
5
0
2
20
9
9
0
0
10
6.
5
0
89
83
0
100
100
21
50
0 ++
o/o
o/o
56
55
46
o/o
APPENDIX. continued
Computer search retrieval
71.
72.
73.
711.
75.
77.
79.
80.
87.
99.
Constant voltage transformers
Manufacture of incandescentand fluorescent lamps
Geothermal power supplies
Use of wood poles for carryingehu transmission lines
Mass soldering methods used inthe production of printedcircuit boards especially dipand flow methods
On-line computers in trafficcontrol (systems in Glasgow,London and Toyko)
Autopilots for ships
Automatic control of vulcanisers
Material properties of GaAs
Thyristor controlled d.c.machines
No. ofreferencesretrieved
(a)
No. selectedby inform-
ationofficer (b)
Precisiona/b
5
8
9
0
0
13
0
It2
5
6
8
0
0
9
0
19
100
75
89
o/o
0/0
69
o/o
if 5
The output of this search was seconded for a seperate projector invalidated by a procedural error.
++ Where the output was of the orderer 200 or 300 references theprecision was estimated on the basis of a random selection fromthe output.
APPENDIX 5
Summary of reasons for retrieval failuresIn the computer search
1. Profile contributed failures
(i) Too many profile fields
(ii) Repetition of concepts in different fields
(iii) Use of 'vague terms' e.g., technological trends
(iv) Use of comparative terms and value rangese.g. small scale.
(v) Insufficient synonyms
(vi) Misuse of truncation and exact matching
(vii). Spelling errors
2. Document indexing
(i) Superficial indexing of books and reviews
(ii)' Indexing by different attributes
APPENDIX 6
Detailed discussion of the exhaustivity of profile fields
Exhaustive profiles as a cause of profile failure.
A common factor in many of the search profiles whichfailed to retrieve was a larger number of profile fieldsi.e. four or more fields which must be satisfied. As thenumber of profile fields increases the probability of effectinga match decreases substantially and if any references are tobe retrieved the number of fields must be kept to a minimum.
Minimising the number of profile fields.
A more detailed consideration of the longer profiles revealed thatmany were unnecessarily long. The occured most commonly wheretwo fields contend synonymous or hierarchically related terms.For example, a que.:T concerning the Lincompex radio-navigationsystem was expressed by a profile which included the twodistinct fields 'Lincompex.' and 'radio-navigation'. SinceLincompex is itself a radio-navigation system it is unnecessaryto include the latter profile field and the shorter profilewould not lack precision. Duplication of this type is moreeasily avoided when a controlled language is used mainlybecause the profile compiler has to hand a structured list of terms.A similar structured list would be most useful for reducing theseerrors. Where a free-language base is being searched.
A more complex variation on this problem arises when a singleprofile field ccrveys-a concept which is already expressed in thecombination of two or more of the remaining fields. Againthis is an occurance particular to the free language vocabularywhich could be reduced if the compiler has access to a structureddisplay. In this system however, the compiler, although aware ofthe problem, may be unable to correct it because of the restrictednature of the Boolean logic and very complex profiles or a moreflexible logic formulation are the only solutions.
The INSPEC database provides the user with a choice of searchfields whose structure differs but whose subject information contentoverlaps, e.g. classification codes and free-indexing terms.Experience has shown that precision in retrieval is achieved byutilising the calssification codes to select a subset of thedata base in which the free indexed profile terms are sought.There are however, occasions when the classification codes canplay a more positive role in the profile compilation. In anumber of queries some or all of the concepts are preciselyexpressed in a classification code and further definition byprofile terms is unnecessary. For example, the concept of'speed control' is defined by the code 73.22 (Control of variables/speed) and the two additional profile fields 'speed' and 'control'
lengthen the profile to no purpose. Where two overlappingsearch fields are available it is important that their useis complementary.
One further contributory factor in the longer profiles wasthe use of profile terms which are implied by the natureof the data base searched. For example, it might reasonablybe expected that devices described in records retrieved fromSection 13 of the INSPEC data base would be either electricalor electronic and not mechanical. To specify these terms istherefore unnecessary. When a single data base is used thispoint is trival but as services are extended to cover a widervariety of the data bases available it will become more important.
APPENDIX 7
Summary of reasons for retrieval of non-relevantitems from computer search.
1. In the statement of the query
(i) Inadequately expressed
2. In the profile
(i) A term has a different meaning in the contextof the document retrieved
(ii) A term has a different meaning or function in thecontext of the indexing phrase from which it isretrieved
(iii) The Boolean operators do not adequately express therelationships between terms
(iv) Not all the concepts of the query are expressed inthe profile
(v) A spelling error
(vi) An error in .a classification code
(vii) Truncated terms retrieve unexpected words
3. In the document indexing
(i) Spelling errors
(ii) Exhaustive indexing
(iii) Misrepresentation of the document
APPENDIX 8
A Summary of the Criteria Used by the InformationOfficer in the Slection of Items Suitable for theEnquirer
1. In screening the papers which accurately satisfy thesubject matter of the enquiry.(These criteria are drawn from the analysis of theinformation officer's assessment of the output of boththe computer and manual searches)
(i) Language
(ii) Level of treatment, e.g. studenttext or veryadvanced
(iii) Bias, e.g. applications, equipment or theory
(iv) Age of paper
(v) Country of origin
(vi) Organisation in which the work was carried out
(vii) Bibliographic form, e.g. journal article, reportor book
(viii) The availability of the full text
(ix) The total number of items retrieved
(x) Standing of the journal cited
2. In selecting papers which are suitable for the enquirerbut do not accurately satisfy the subject matter ofthequery.(These criteria are drawn from the analysis of the manualsearches)
(i) Reduction in the exhaustivity of the query
(ii) Associated subjects or techniques
(iii) Indexing breakdown by different attributes
(iv) Synonymous terms
(v) Reduced or increased specificity of terms
APPENDIX 9
Suggested criteria for queries best suited to abatch - processed computer search or a manual search
1. For a batch-processed computer search
(i) Tf an exhaustive search for the compilation ofa bibliography is required
(ii) If the limited time span of the data base issufficient
(iii) If a single concrete term is sought which possiblycuts across the classification and is not a subject-index heading.
2. For a manual search
(i) If the query can be expressed by a subject indexheading and a single qualifier.
(ii) A query which is subjective in nature e.g.lrecentadvances'
(iii) A query which implies a large number of alternativeswhich cannot easily be listed e.g. 'Electric handtools'.
(iv) A query for basic information which are best answeredby books.
(v) Where range data is required e.g. over 10,000 rpm,small scale.
APPENDIX 10
ciuery information for successful profile compilation
(i) Subject matter - clearly defined
(ii) The use which is to be made of the references,e.g. to write a review for a popular magazine orto compile a complete bibliography.
(iii) An approximation of the size of output which wouldbest please the enquirer.
(iv) The level of treatment which the enquirer requiresor can understand, e.g. student text or highlyadvanced.
(v) Any special applications or bias of treatment, e.g.theory or equipment.
(vi) Languages which the enquirer can read or readilyhave translated.
(vii) Details of the approximate period when the work in thesubject field was carried out.
(viii) Any restrictions on the dates of publications retrieved.
(ix) Preferences for country or organisation of origin ofitems.
(x) If time is an important factor, the journals and typeof literature from the full-text is available areimportant.
Guidelines for tape use and profile compilation
A. Notes on the INSPEC data base
(±)
APPENDIX 11
English language items are not usuallystated as such. If searches are requires on thisdata a search for an empty field is necessary.Alternatively the field may be filled airomaticallybefore processing.
(ii) Treatment codes were not included on the recordson the early tapes. Treatment codes must not be'necessary' to the retrieval if documents from theolder tapes are desired.
(iii) The indexing of the data base does not supportexhaustive profiles with many 'necessary' fields.A match on four or more fields is unusual.
(iv) The indexing of book material is not always sufficientlydetailed for books and journal papers to be soughtusing the same profile.
(v) The use of multiword profile terms does not geaerallyproduce high recall.
B. Notes on profile construction
(±) It is essential that the profile compiler has suchinformation about the enquiry as is necessary(Appendix 9)
The data base must have a suitab;e subject coverageand cover a suitable time period for the query e.g.the majority of work in 4 particular field may haveto he carried oct before the commencement of thedata base.
(iii) Care should be taken that a query which representstwo approaches to a problem is expressed as twoprofiles or separate parts of a profile.
(iv) Subjective profile terms such as 'developments' and'applications' are rarely successful.
(v) The use of comparative terms e.g. vhf and numericalranges in a profile is not usually successful.
(vi) Care should be taken that a profile does not requirea match on a single concept in more than one field.
