1. 1 Query Language Baeza-Yates and Navarro Modern Information Retrieval, 1999 Chapter 4

2. 2 Query Language l Keyword-based Querying »Single-word Queries »Context Queries –Phrase –Proximity »Boolean Queries »Natural Language

3. 3 Query Language (Cont.) l Pattern Matching »Words »Prefixes »Suffixes »Substring »Ranges »Allowing errors »Regular expressions

4. 4 Query Language (Cont.) l Structural Queries »Form-like fixed structures »Hypertext structure »hierarchical structure

5. 5 Structural Queries (a) form-like fixed structure, (b) hypertext structure, and (c) hierarchical structure

6. 6 Hierarchical Structure An example of a hierarchical structure: the page of a book, its schematic view, and a parsed query to retrieve the figure

7. 7 The Types of Queries Boolean queries fuzzy Boolean Natural language structural queriesbasic queries proximity phrases pattern matching errors words substringsregular expressions keywords and context prefixesextended patterns suffixes

8. 8 Query Operations Baeza-Yates, 1999 Modern Information Retrieval Chapter 5

9. 9 Query Modification l Improving initial query formulation »Relevance feedback –approaches based on feedback information from users »Local analysis –approaches based on information derived from the set of documents initially retrieved (called the local set of documents) »Global analysis –approaches based on global information derived from the document collection

10. 10 Relevance Feedback l Relevance feedback process »it shields the user from the details of the query reformulation process »it breaks down the whole searching task into a sequence of small steps which are easier to grasp »it provides a controlled process designed to emphasize some terms and de-emphasize others l Two basic techniques »Query expansion –addition of new terms from relevant documents »Term reweighting –modification of term weights based on the user relevance judgement

11. 11 Vector Space Model l Definition w i,j : the i th term in the vector for document d j w i,k : the i th term in the vector for query q k t: the number of unique terms in the data set

12. 12 Query Expansion and and Term Reweighting for the Vector Model l Ideal situation »C R : set of relevant documents among all documents in the collection l Rocchio (1965, 1971) »R: set of relevant documents, as identified by the user among the retrieved documents »S: set of non-relevant documents among the retrieved documents

13. 13 Rocchio’s Algorithm l Ide_Regular (1971) l Ide_Dec_Hi l Parameters  =  =  =1  > 

14. 14 Probabilistic Model l Definition »p i : the probability of observing term t i in the set of relevant documents »q i : the probability of observing term t i in the set of nonrelevant documents l Initial search assumption »p i is constant for all terms t i (typically 0.5) »q i can be approximated by the distribution of t i in the whole collection

15. 15 Term Reweighting for the Probabilistic Model l Robertson and Sparck Jones (1976) l With relevance feedback from user N: the number of documents in the collection R: the number of relevant documents for query q n i : the number of documents having term t i r i : the number of relevant documents having term t i Document Relevance Document Indexing + - + riri R-r i R N-n i -R+r i - n i -r i N-R nini N-n i N

16. 16 l Initial search assumption »p i is constant for all terms t i (typically 0.5) »q i can be approximated by the distribution of t i in the whole collection l With relevance feedback from users »p i and q i can be approximated by »hence the term weight is updated by Term Reweighting for the Probabilistic Model (cont.)

17. 17 l However, the last formula poses problems for certain small values of R and r i (R=1, r i =0) l Instead of 0.5, alternative adjustments have been propsed Term Reweighting for the Probabilistic Model (Cont.)

18. 18 l Characteristics »Advantage –the term reweighting is optimal under the asumptions of l term independence l binary document indexing (w i,q  {0,1} and w i,j  {0,1}) »Disadvantage –no query expansion is used –weights of terms in the previous query formulations are also disregarded –document term weights are not taken into account during the feedback loop Term Reweighting for the Probabilistic Model (Cont.)

19. 19 Evaluation of relevance feedback l Standard evaluation method is not suitable »(i.e., recall-precision) because the relevant documents used to reweight the query terms are moved to higher ranks. l The residual collection method »the set of all documents minus the set of feedback documents provided by the user »because highly ranked documents are removed from the collection, the recall-precision figures for tend to be lower than the figures for the original query »as a basic rule of thumb, any experimentation involving relevance feedback strategies should always evaluate recall- precision figures relative to the residual collection

20. 20 Automatic Local Analysis l Definition »local document set D l : the set of documents retrieved by a query »local vocabulary V l : the set of all distinct words in D l »stemed vocabulary S l : the set of all distinct stems derived from V l l Building local clusters »association clusters »metric clusters »scalar clusters

21. 21 Association Clusters l Idea »co-occurrence of stems (or terms) inside documents –f u,j : the frequency of a stem ku in a document d j »local association cluster for a stem k u –the set of k largest values c(k u, k v ) »given a query q, find clusters for the |q| query terms »normalized form

22. 22 Metric Clusters l Idea »consider the distance between two terms in the same cluster l Definition »V(k u ): the set of keywords which have the same stem form as k u »distance r(k i, k j )=the number of words between term k u and k v »normalized form

23. 23 Scalar Clusters l Idea »two stems with similar neighborhoods have some synonymity relationships l Definition »c u,v =c(k u, k v ) »vectors of correlation values for stem k u and k v »scalar association matrix »scalar clusters –the set of k largest values of scalar association

24. 24 Automatic Global Analysis l A thesaurus-like structure l Short history »Until the beginning of the 1990s, global analysis was considered to be a technique which failed to yield consistent improvements in retrieval performance with general collections »This perception has changed with the appearance of modern procedures for global analysis

25. 25 Query Expansion based on a Similarity Thesaurus l Idea by Qiu and Frei [1993] »Similarity thesaurus is based on term to term relationships rather than on a matrix of co-occurrence »Terms for expansion are selected based on their similarity to the whole query rather than on their similarities to individual query terms l Definition »N: total number of documents in the collection »t: total number of terms in the collection »tf i,j : occurrence frequency of term k i in the document d j »t j : the number of distinct index terms in the document d j »itf j : the inverse term frequency for document d j

26. 26 Similarity Thesaurus l Each term is associated with a vector »where w i,j is a weight associated to the index-document pair l The relationship between two terms k u and k v is »Note that this is a variation of the correlation measure used for computing scalar association matrices

27. 27 Term weighting vs. Term concept space tf ij Term k i Doc d j tf ij Term k i Doc d j

28. 28 Query Expansion Procedure with Similarity Thesaurus 1. Represent the query in the concept space by using the representation of the index terms 2. Compute the similarity sim(q,k v ) between each term k v and the whole query 3. Expand the query with the top r ranked terms according to sim(q,k v )

29. 29 Example of Similarity Thesaurus The distance of a given term k v to the query centroid Q C might be quite distinct from the distances of k v to the individual query terms kaka kbkb kiki kjkj kvkv QCQC Q C ={k a,k b }

30. 30 Query Expansion based on a Similarity Thesaurus »A document d j is represented term-concept space by »If the original query q is expanded to include all the t index terms, then the similarity sim(q, d j ) between the document d j and the query q can be computed as –which is similar to the generalized vector space model

31. 31 Query Expansion based on a Statistical Thesaurus l Idea by Crouch and Yang (1992) »Use complete link algorithm to produce small and tight clusters »Use term discrimination value to select terms for entry into a particular thesaurus class l Term discrimination value »A measure of the change in space separation which occurs when a given term is assigned to the document collection

32. 32 Term Discrimination Value l Terms »good discriminators: (terms with positive discrimination values) –index terms »indifferent discriminators: (near-zero discrimination values) –thesaurus class »poor discriminators: (negative discrimination values) –term phrases l Document frequency df k »df k >n/10: high frequency term (poor discriminators) »df k <n/100: low frequency term (indifferent discriminators) »n/100  df k  n/10: good discriminator

33. 33 Statistical Thesaurus l Term discrimination value theory »the terms which make up a thesaurus class must be indifferent discriminators l The proposed approach »cluster the document collection into small, tight clusters »A thesaurus class is defined as the intersection of all the low frequency terms in that cluster »documents are indexed by the thesaurus classes »the thesaurus classes are weighted by

34. 34 Discussion l Query expansion »useful »little explored technique l Trends and research issues »The combination of local analysis, global analysis, visual displays, and interactive interfaces is also a current and important research problem