Information retrieval (IR) Basics, models, interactions tefkos@rutgers.edu; http://comminfo.rutgers.edu/~tefko/ Tefko Saracevic 1 Central ideas • Information retrieval (IR) is at the heart of ALL indexing & abstracting databases, information resources, and search engines – all work on basis of IR algorithms and procedures • Contemporary IR is also interactive – to such a degree that pragmatically IR can not be separated from interaction • As a searcher you will constantly use IR, thus you have to be knowledgeable about it Tefko Saracevic 2 ToC 1. 2. 3. 4. Information retrieval (IR) Matching algorithms: Exact match & best match Strength & weaknesses IR Interaction & interactive models Tefko Saracevic 3 1. Information retrieval Definitions. Traditional model Tefko Saracevic 4 Information retrieval (IR) - original definition Calvin Mooers (1919-1994) coined the term “Information retrieval embraces the intellectual aspects of the description of information and its specification for search, and also whatever systems, techniques, or machines are employed to carry out the operation.” Mooers, 1951 Tefko Saracevic 5 IR: Objective & problems Objectives: Provide users with effective access to & interaction with information resources. Retrieve information or information objects that are relevant Problems addressed: 1. How to organize information intellectually? 2. How to specify search & interaction intellectually? 3. What systems & techniques to use for those processes? Tefko Saracevic 6 IR models • Model depicts, represents what is involved - a choice of features, processes, things for consideration • Several IR models used over time – traditional: oldest, most used, shows basic elements involved – interactive: more realistic, favored now, shows also interactions involved; several models proposed • Each has strengths, weaknesses • We start with traditional model to illustrate many points - from general to specific examples Tefko Saracevic 7 Description of traditional IR model • It has two streams of activities – one is the systems side with processes performed by the system – other is the user side with processes performed by users & intermediaries (you) – these two sides led to “system orientation” & “user orientation” – in system side automatic processing is done; in user side human processing is done • They meet at the matching process – where the query is fed into the system and system looks for documents that match the query • Also feedback is involved so that things change based on results – e.g. query is modified & new matching done Tefko Saracevic 8 Traditional IR model System Acquisition Problem documents, objects information need Representation Representation indexing, ... question File organization Query indexed documents search formulation Matching searching feedback User Retrieved objects Tefko Saracevic 9 Acquisition system side • Content: What is in databases – In Dialog first part of blue sheets: File Description, Subject Coverage; in Scopus Subject Areas • Selection of documents & other objects from various sources - journals, reports … – In Blue Sheets: Sources; in Scopus Sources • Mostly text based documents – Full texts, titles, abstracts ... – But also: data, statistics, images (e.g. maps, trade marks) ... Importance: Determines contents of databases Key to file selection in searching !!! Tefko Saracevic 10 Representation of documents, objects … system side • Indexing – many ways : – free text terms (even in full texts) – controlled vocabulary thesaurus – manual & automatic techniques • Abstracting; summarizing • Bibliographic description: – author, title, sources, date… – metadata • Classifying, clustering • Organizing in fields & limits – in Dialog: Basic Index, Additional Index. Limits – in Scopus pull down menus Basic to what is available for searching & displaying Tefko Saracevic 11 File organization system side As mentioned: • Sequential – record (document) by record • Inverted – term by term; list of records under each term • Combination: indexes inverted, documents sequential • When citation retrieved only, need for document files or document delivery Enables searching & interplay between types of files Tefko Saracevic 12 Problem user side • Related to user’s task, situation, problem at hand – vary in specificity, clarity • Produces information need – ultimate criterion for effectiveness of retrieval • how well was the need met? • Inf. need for the same problem may change, evolve, shift during the IR process - adjustment in searching – often more than one search for same problem over time • you will experience this in your term project Critical for examination in interview Tefko Saracevic 13 Representation question – user side Non-mediated: end user alone Mediated: intermediary + user – interviews; humanhuman interaction • Question analysis – selection, elaboration of terms – various tools may be used • thesaurus, classification schemes, dictionaries, textbooks, catalogs … • Focus toward – deriving search terms & logic – selection of files, resources • Subject to feedback changes • Critical roles of intermediary - you Determines search specification - a dynamic process Tefko Saracevic 14 Query search formulation – user side • Translation into systems requirements & limits – start of human-computer interaction • Selection of files, resources • Search strategy - selection of: – search terms & logic – possible fields, delimiters – controlled & uncontrolled vocabulary – variations in tactics • Reiterations from feedback – several feedback types: relevance feedback, magnitude feedback ... – query expansion & modification What & how of actual searching Tefko Saracevic 15 Matching searching – system side • Process of comparing – search: what documents in the file match the query as stated? • Various search algorithms: – exact match - Boolean • Each has strengths, weaknesses – no ‘perfect’ method exists • and probably never will • still available in most, if not all systems – best match - ranking by relevance • increasingly used e.g. on the web – hybrids incorporating both • e.g. Target, Rank in Dialog Involves many types of search interactions & formulations Tefko Saracevic 16 Retrieved objects from system to user • Various order of output: • When citations only – sorted by Last In First Out available: possible links (LIFO) to document delivery – ranked by relevance & then LIFO – ranked by other characteristics • Various forms of output – In Dialog: Output options – in Scopus title (default), abstract + references, cited by, plus more – Scopus View at publisher – accessing RUL for digital journals • Base for relevance, utility evaluation by users What a user (or you) sees, gets, judges – can be specified Tefko Saracevic 17 2. Matching algorithms Exact match & best match searches Tefko Saracevic 18 Exact match - Boolean search • You retrieve exactly what you ask for in the query: – all documents that have the term(s) with logical connection(s), and possible other restrictions (e.g. to be in titles) as stated in the query – exactly: nothing less, nothing more • Based on matching following rules of Boolean algebra, or algebra of sets – ‘new algebra’ – presented by circles in Venn diagrams Tefko Saracevic 19 Boolean algebra: operates on sets of documents • Has four operations (like in algebra): 1. A: retrieves set that has term A • • Tefko Saracevic • • I want documents that have the term library 2. A AND B: retrieves set that has terms A and B • 3. A OR B: retrieves set that has either term A or B often called intersection & labeled A B I want documents that have both terms library and digital someplace within often called union and labeled A B I want documents that have either term library or term digital someplace within 4. A NOT B: retrieves set that has term A but not B • • often called negation and labeled A – B I want documents that have term library but if they also have term digital I do not want those 20 Potential problems • But beware: – digital AND library will retrieve documents that have digital library (together as a phrase) but also documents that have digital in the first paragraph and library in the third section, 5 pages later, and it does not deal with digital libraries at all • thus in Scopus or Google you will ask for “digital library” and in Dialog for digital(w)library to retrieve the exact phrase digital library – digital NOT library will retrieve documents that have digital and suppress those that along with digital also have library, but sometimes those suppressed may very well be relevant. Thus, NOT is also known as the “dangerous operator “ – also beware of order: venetian AND blind will retrieve documents that have venetian blind and also that have blind venetian (oldest joke in information retrieval) Tefko Saracevic 21 Boolean algebra depicted in Venn diagrams Four basic operations: e.g. A = digital B= libraries A B 1 2 3 A 1 B 2 A 1 Tefko Saracevic 3 A AND B. Shade 2 digital AND libraries B 2 A 1 A alone. All documents that have A. Shade 1 & 2. digital 3 A OR B. Shade 1, 2, 3 digital OR libraries B 2 3 A NOT B. Shade 1 digital NOT libraries 22 Venn diagrams … cont. Complex statements allowed e.g A B 2 1 4 5 3 6 (A OR B) AND C Shade 4,5,6 (digital OR libraries) AND Rutgers 7 C Tefko Saracevic (A OR B) NOT C Shade what? (digital OR libraries) NOT Rutgers 23 Venn diagrams cont. • Complex statements can be made – as in ordinary algebra e.g. (2+3)x4 • As in ordinary algebra: watch for parenthesis: – 2+(3 x 4) is not the same as (2+3)x4 – (A AND B) OR C is not the same as A AND (B OR C) Tefko Saracevic 24 Adding variations to Boolean searches • digital AND libraries can be specified to appear in given fields as present in the given system – e.g. to appear in titles only • in Dialog command is s digital AND libraries/TI • in Scopus pull down menu allows for selection of given field, – so for digital library specify Article Title in pull down menu • in Google Advanced Search gets you to a pull down menu for Where your keywords show up: & then go to in the title of the page • Various systems have different ways to retrieve singular and plurals for the same term • in Scopus term library will retrieve also libraries & vice versa • in Dialog you have to specify librar? to retrieve variants • in Google library retrieves library but not libraries Tefko Saracevic 25 Best match searching • Output is ranked – it is NOT presented as a Boolean set but in some rank order • You retrieve documents ranked by how similar (close) they are to a query (as calculated by the system) – similarity assumed as relevance – ranked from highest to lowest relevance to the query • mind you, as considered by the system • you change the query, system changes rank – thus, documents as answers are presented from those that are most likely relevant downwards to less & less likely relevant as determined by a given algortihm – remember: a system algorithm determines relevance ranking Tefko Saracevic 26 Best match ... cont. • Best match process deals with PROBABILITY: – – – – – • what is the probability that a document is relevant to a query? compares the set of query terms with the sets of terms in documents calculates a similarity between query & each document based on common terms &/or other aspects sorts the documents in order of similarity assumes that the higher ranked documents have a higher probability of being relevant allows for cut-off at a chosen number e.g. the first 20 documents • BIG issue: What representation & similarity measures are better? Subject of IR experiments – “better” determined by a number of criteria, e.g. relevance, speed … Tefko Saracevic 27 Best match (cont.) • Variety of algorithms (formulas) used to determine similarity – using statistic &/or linguistic properties • e.g. if digital appears a lot of times in a given document relative to its size, that document will be ranked higher when the query is digital – many proposed & tested in IR research – many developed by commercial organizations • Google also uses calculations as to number of links to/from a document & other methods • many algorithms are now proprietary & not disclosed – the way a system ranks and you rank may not necessarily be in agreement • Web outputs are mostly ranked – but Dialog allows ranking as well, with special commands Tefko Saracevic 28 3. Strengths & weaknesses Best vs. exact match Traditional IR model Tefko Saracevic 29 Boolean vs. best match • Boolean – allows for logic – provides all that has been matched BUT – has no particular order of output – usually LIFO – treats all retrievals equally from the most to least relevant ones – often requires examination of large outputs Tefko Saracevic • Best match – allows for free terminology – provides for a ranked output – provides for cut-off - any size output BUT – does not include logic – ranking method (algorithm) not transparent • whose relevance? – where to cut off? 30 Strengths of traditional IR model • Lists major components in both system & user branches • Suggests: – What to explain to users about system, if needed – What to ask of users for more effective searching (problem ...) • Aids in selection of component(s) for concentration – mostly ever better representation • Provides a framework for evaluation of (static) aspects Tefko Saracevic 31 Weaknesses • Does not address nor account for interaction & judgment of results by users – identifies interaction with matching only – interaction is a much richer process • Many types of & variables in interaction not reflected • Feedback has many types & functions - also not shown • Evaluation thus one-sided IR is a highly interactive process - thus additional model(s) needed Tefko Saracevic 32 4. IR interaction Models. Implications: what happens in searching? Tefko Saracevic 33 Enters interaction There is MUCH more to searching than knowing computers, networks & commands, as there is more to writing than knowing word processing packages Tefko Saracevic 34 IR as interaction • If we consider USER & USE central, then: Interaction is a dominant feature of contemporary IR • Interaction has many facets: – with systems, technology – with documents, texts viewed/retrieved – intermediaries with people • Several interactive IR models – none as widely accepted as traditional IR model • Broader area: human-computer interaction (HCI) studies Tefko Saracevic 35 HCI: broader concepts “Any interaction takes place through one or more interfaces & involves two or more participants who each have one or more purposes for the interaction” Storrs, 1994 • Participants: people & ‘computer’ (everything in it – software, hardware, resources …) • Interface: a common boundary • Purposes: people have purposes and ‘computer’ has purposes built in • At issue: identification of important aspects, roles of each Tefko Saracevic 36 HCI … definitions “Interaction is the exchange of information between participants where each has the purpose of using the exchange to change the state of itself or of one or more of others” “An interaction is a dialogue for the purpose of modifying the state of one or more participants” • Key concepts: exchange, change – for user: change the state of knowledge related to a given problem, tasks, situation Tefko Saracevic 37 IR interaction is ... “... the interactive communication processes that occur during the retrieval of information by involving all the major participants in IR, i.e. the user, the intermediary, and the IR system.” Ingwersen, 1992 • Involved: – users – intermediaries (possibly) – everything in IR system – communication processes - exchange of information Tefko Saracevic 38 Questions • What variables are involved in interaction? – models give lists • How do they affect the process? How to control? – experiments, experience, observation give answers • Do given interventions (actions) or communications improve or degrade the process? – e.g. searcher’s (intermediaries or end-users) actions • Can systems be designed so that searcher’s intervention improves performance? Tefko Saracevic 39 Interactive IR models • Several models proposed – none as widely accepted as the traditional IR model • They all try to incorporate – – – – – information objects (“texts”): IR system & setting interface intermediary, if present user’s characteristics • cognitive aspects; task; problem; interest; goal; preferences ... – social environment – variety of processes between them all. Tefko Saracevic 40 User modeling (treated in unit 11, but introduced here to illustrate one of the important aspect of human-human interaction) • Identifying elements about a user that impact interaction, searching, types of retrieval …: – – – – – – – who is the user (e.g. education) what is the problem, task at hand what is the need; question how much s/he knows about it what will be used for how much wanted, how fast what environment is involved • Much more than just analyzing a question posed by user – related to reference interview • Used to select resources, specify search concepts and terms, formulate query, select format and amount of results provided, follow up with feedback and reiteration, change tactics … Tefko Saracevic 41 Three interactive models • Three differing models are presented here, each concentrates on a different thing: – Ingwersen concentrates on enumeration of general elements that enter in interaction – Belkin on different processes that are involved as interaction progresses through time – Saracevic on strata or levels of interaction elements on computer and user side • As mentioned, no one interaction model is widely accepted as the traditional IR model Tefko Saracevic 42 Ingwersen’s interactive cognitive model • Among the first to view IR differently from traditional model • Included IR as a system but concentrates also on elements outside system that interact – – – – – inf. objects – documents, images … intermediary – you - & interface user cognitive aspects user & general environment path of request (we call question) • from environment (problem) to query – path of cognitive changes – path of communication – various other paths of interactions Tefko Saracevic 43 Ingwersen’s model graphically Information objects Interface/ Intermediary Query IR system setting User’s cognitive space Environ ment Request Cognitive transformations Interactive communication Tefko Saracevic 44 Belkin’s episodes model • Concentrates on what happen in interaction as process – Ingwerson concentrated on elements • Viewed interaction as a series of episodes where a number of different things happen over time – depending on user’s goals, tasks • there is judgment, use, interpretation… – processes of navigation, comparison, summarization … – involving different aspects of information & inf. objects • While interacting we do diverse things, perform various tasks, & involve different objects Think: what do you do while searching? Tefko Saracevic 45 Belkin’s episodes model USER USER USER CO CO COMPARISON Goals tasks REPRESENTATION ..... INTERACTION Judgment, use, interpretation, modification SUMMARIZATION NA NA NAVIGATION INFORMATION VISUALIZATION Type, medium mode level Tefko Saracevic 46 Saracevic’ stratified model • Interaction: considers it as a sequence of processes/episodes occurring in several levels or strata* Interaction = INTREPLAY between levels • Structure: – – – – – Several User levels Produce a Query – it has characteristics Several Computer levels They all meet on the Surface level Dialogue enabled by Interface • user utterances • computer ‘utterances’ • Adaptation/changes in all • Geared toward Information use Tefko Saracevic 47 Saracevic’s stratification model Context social, cultural … Situational tasks; work context... Affective intent; motivation ... Cognitive knowledge; structure... Query characteristics … Surface level INTERFACE Engineering hardware; connections... Processing software; algorithms … Content inf. objects; representations... Tefko Saracevic 48 Roles of levels or strata • Defining of what’s involved – whassup? • Help in recognition/separation of differing variables – each strata or level involves different elements, roles, & processes • Observation of interaction between strata complex dynamics • On the user side suggests what affects factors query and judgment of responses – thus elements for user modeling Tefko Saracevic 49 Interplay between levels • Interplay on user side: – Cognitive: between cognitive structures of texts & users – Affective: between intentions & other – Situational: between texts & tasks • Similar interplay on computer side • Surface level - interface: – searching, navigation, browsing, display, visualization, query characterization … • Interplay judgments in searching: – – – – evaluation of results - relevance changing of models: situation, need ... selection of search terms resulting modifications - feedback Tefko Saracevic 50 Intermediaries - YOU • Intermediaries could participate as an additional interface - many roles: – diagnostic help in problem, query formulation – system interface handling – selection, interpretation & manipulation of inf. resources – interpretation of results – education of users – enablers of end-users • Basic role: optimizing results • Act in processes at different levels Tefko Saracevic 51 Implications • Interaction central to IR including in searching of the Web • We see it on the surface level – But result of MANY variables, levels & their interplay • IR interaction requires knowledge of these levels & interplays – many users have difficulties – so do many professionals • Design of interfaces for interaction still lacking • People compensate in many ways including trial & error, failures Tefko Saracevic 52 What happens in searching? • Highly reiterative process – back & forth between user modeling & (re)formulating search strategy – goes on & on in many feedback loops, twists & turns, shifts • Search strategy (the big picture) – selection/reselection of sources – stating a query (search statement) from a question • terms, their expansions, logic, qualifications, limitations Tefko Saracevic Searching … (cont.) • Search tactics (action steps) – what to do first, next – e.g. from broad to narrow searches – format of results • Evaluation of results – as to magnitude - how much? – as to relevance - how well? – feedback to change after that • user model - e.g. question • strategy - e.g. files, query • tactics - e.g. narrowing, broadening Tefko Saracevic 54 Practical suggestions for searchers (filched from a source I cannot find anymore) • Prepare carefully • Understand your opponent – e.g. Dialog, Scopus, LexisNexis • Anticipate – e.g. hidden meaning of terms • Have a contingency plan – assessing odds of success or points of diminishing returns • Avoid ambiguity – inherent in language • Stay loose! Tefko Saracevic 55 Stay loose? • I copied that, but always wandered what does it really mean? • Dictionary says: not firmly fastened or fixed in place • ???? well, sounds OK! • or Tefko Saracevic 56 Tefko Saracevic 57