Information Retrieval (IR) Deals with the representation, storage and retrieval of unstructured data Topics of interest: systems, languages, retrieval, user interfaces, data visualization, distributed data sets Classical IR deals mainly with text The evolution of multimedia databases and of the web have given new interest to IR E.G.M Petrakis Multimedia Information Retrieval 1 Multimedia IR A multimedia information system that can store and retrieve attributes text 2D grey-scale and color images 1D time series digitized voice or music video E.G.M Petrakis Multimedia Information Retrieval 2 Applications Financial, marketing: stock prices, sales etc. find companies whose stock prices move similarly Scientific databases: sensor data whether, geological, environmental data Office automation, electronic encyclopedias, electronic books Medical databases X-rays, CT, MRI scans Criminal investigation suspects, fingerprints, Personal archives, text and color images E.G.M Petrakis Multimedia Information Retrieval 3 Queries Formulation of user’s information need Free-text query By example document (e.g., text, image) e.g., in a collection of color photos find those showing a tree close to a house Retrieval is based on the understanding of the content of documents and of their components E.G.M Petrakis Multimedia Information Retrieval 4 Goals of Retrieval Accuracy: retrieve documents that the user expects in the answer With as few incorrect answers as possible All relevant answers are retrieved Speed: retrievals has to be fast The system responds in real time E.G.M Petrakis Multimedia Information Retrieval 5 Accuracy of Retrieval Depends on query criteria, query complexity and specificity Attributes / Content criteria? Depends on what is matched with the query How documents content is represented Typically feature vectors Depends also on matching function Euclidean distance E.G.M Petrakis Multimedia Information Retrieval 6 Speed of Retrieval The query is compared (matched) with all stored documents The definition of similarity criteria (similarity or distance function between documents) is an important issue: Similarity or distance function between documents Matching has to fast: document matching has to be computationally efficient and sequential searching must be avoided Indexing: search documents that are likely to match the query E.G.M Petrakis Multimedia Information Retrieval 7 Database index query database descriptions documents Doc1-FeactureVector Doc2-FeactureVector Doc1 Doc2 DocN DocN-FeatureVector E.G.M Petrakis Multimedia Information Retrieval 8 Main Idea Descriptions are extracted and stored Same or separate storage with documents Queries address the stored descriptions rather the documents themselves Images & video: the majority of stored data Solution: retrieval by text Text retrieval is a well researched area Most systems work with text, attributes E.G.M Petrakis Multimedia Information Retrieval 9 Problem Text descriptions for images and video contained in documents are not available e.g., the text in a web site is not always descriptive of every particular image contained in the web site Two approaches human annotations feature extraction E.G.M Petrakis Multimedia Information Retrieval 10 Human Annotations Humans insert attributes, captions for images and videos inconsistent or subjective over time and users (different users do not give the same descriptions) retrievals will fail if queries are formulated using different keywords or descriptions time consuming process, expensive or impossible for large databases E.G.M Petrakis Multimedia Information Retrieval 11 Feature Extraction Features are extracted from audio, image and video cheaper and replicable approach consistent descriptions but inexact low level feature (patterns, colors etc.) difficult to extract meaning different techniques for different data E.G.M Petrakis Multimedia Information Retrieval 12 Architecture of a IR System for Images Furht at.al. 96 E.G.M Petrakis Multimedia Information Retrieval 13 Multimedia IR in Practice Relies on text descriptions Non-text IR there is nothing similar to text-based retrieval systems automatic IR is sometimes impossible requires human intervention at some level significant progress have been made domain specific interpretations E.G.M Petrakis Multimedia Information Retrieval 14 Ranking of IR Methods Ranking in terms of complexity and accuracy attributes text audio image video accuracy complexity Combined IR based on two or more data types for more accurate retrievals Complex data types (e.g.,video) are rich sources of information E.G.M Petrakis Multimedia Information Retrieval 15 Similarity Searching IR is not an exact process two documents are never identical! they can be “similar” searching must be “approximate” The effectiveness of IR depends on the types and correctness of descriptions used types of queries allowed user uncertainly as to what he is looking for efficiency of search techniques E.G.M Petrakis Multimedia Information Retrieval 16 Approximate IR A query is specified and all documents up to a pre-specified degree of similarity are retrieved and presented to the user ordered by similarity Two common types of similarity queries: range queries: retrieve all documents up to a distance “threshold” T nearest-neighbor queries: retrieve the k best matches E.G.M Petrakis Multimedia Information Retrieval 17 Distance - Similarity Decide whether two documents are similar distance: the lower the distance, the more similar the documents are similarity: the higher the similarity, the more similar the documents are Key issue for successful retrieval: the more accurate the descriptions are, the more accurate the retrieval is E.G.M Petrakis Multimedia Information Retrieval 18 Retrieval Quality Criteria Retrieval with as few errors as possible Two types of errors: false dismissals or misses: qualifying but non retrieved documents false positives or false drops: retrieved but not qualifying documents A good method minimizes both Ranking quality: retrieve qualifying documents before non-qualifying ones E.G.M Petrakis Multimedia Information Retrieval 19 Evaluation of Retrieval “Given a collection of documents, a set of queries and human expert’s responses to the above queries, the ideal system will retrieve exactly what the human dictated” The deviations from the above are measured number of retrieved relevantin answer precision = number of retrieved number or retrieved relevantin answer recall = total number or relevantin the collection E.G.M Petrakis Multimedia Information Retrieval 20 precision Precision-Recall Diagram 1 the ideal method 0,5 1 0,5 recall Measure precision-recall for 1, 2 ..N answers High precision means few false alarms High recall mean few false dismissals E.G.M Petrakis Multimedia Information Retrieval 21 Harmonic Mean A single measure combining precision and recall F = 2 1 +p 1 r: recall, p: precision r F takes values in [0,1] F 1 as more retrieved documents are relevant F 0 as few retrieved documents are relevant F is high when both precision and recall are high F expresses a compromise between precision and recall E.G.M Petrakis Multimedia Information Retrieval 22 Ranking Quality The higher the Rnorm the better the ability of a method to retrieve correct answer before incorrect S S 1 2 (1 ) if S max 0 S max Rnorm 1 otherwise A human expert evaluates the answer set Then take answers in pairs: (relevant,irrelevant) S+ : pairs ranked correctly (the relevant entry has retrieved before the irrelevant one) S- : pairs ranked incorrectly Smax+: total ranked pairs E.G.M Petrakis Multimedia Information Retrieval 23 Searching Method Sequential scanning: the query is matched with all stored documents slow if the database is large or if matching is slow The documents must be indexed hashing, inverted files, B-trees, R-trees different data types are indexed and searched separately E.G.M Petrakis Multimedia Information Retrieval 24 Goals of IR Summarizing, there are two general goals common to all IR systems effectiveness: IR must be accurate (retrieves what the user expects to see in the answer) efficiency: IR must be fast (faster than sequential scanning) E.G.M Petrakis Multimedia Information Retrieval 25 Approximate IR A query is specified and all documents up to a pre-specified degree of similarity are retrieved and presented to the user ordered by similarity Two common types of similarity queries: range queries: retrieve all documents up to a distance “threshold” T nearest-neighbor queries: retrieve the k best matches E.G.M Petrakis Multimedia Information Retrieval 26 Query Formulation Must be flexible and convenient SQL queries constraints on all attributes and data types may become very complex Queries by example e.g., by providing an example document or image Browsing: display headers, summaries, miniatures etc. or for refining the retrieved results E.G.M Petrakis Multimedia Information Retrieval 27 Access Methods for Text Access methods for text are interesting for at least 3 reasons multimedia documents contain text, e.g., images often have captions text retrieval has several applications in itself (library automation, web search etc.) text retrieval research has led to useful ideas like, vector space model, information filtering, relevance feedback etc. E.G.M Petrakis Multimedia Information Retrieval 28 Text Queries Single or multiple keyword queries context queries: phrases, word proximity boolean: keywords with and, or, not Natural language: free text queries Structured search takes also text structure into account and can be: flat or hierarchical for searching in titles, paragraphs, sections, chapters hypertext: combines content-connectivity E.G.M Petrakis Multimedia Information Retrieval 29 Keyword Matching The whole collection is searched no preprocessing no space overhead updates are easy The user specifies a string (regular expression) and the text is parsed using a finite state automaton KMP, BMH algorithms E.G.M Petrakis Multimedia Information Retrieval 30 Error Tolerant Methods Methods that can tolerate errors Scan text one character at a time by keeping track of matched characters and retrieve strings within a desired editing distance from query Wu, Manber 92, Baeza-Yates, Connet 92 Extension: regular expressions built-up by strings and operators: “pro (blem | tein) (s | ε) | (0 | 1 | 2)” E.G.M Petrakis Multimedia Information Retrieval 31 Error Counting Methods Retrieve similar words or phrases e.g., misspelled words or words with different pronunciation editing distance: a numerical estimate of the similarity between 2 strings phonetic coding: search words with similar pronunciation (Soundex, Phonix) N-grams: count common N-length substrings E.G.M Petrakis Multimedia Information Retrieval 32 Editing Distance Minimum number of edit operations that are needed to transform the first string to the second edit operation: insert, delete, substitute and (sometimes) transposition of characters d(si,ti) : distance between characters usually d(si,ti) = 1 if si < > ti and 0 otherwise edit(cordis,codis) = 1: r is deleted edit(cordis, codris) = 1: transposition of rd E.G.M Petrakis Multimedia Information Retrieval 33 j 0 1 2 3 4 5 6 7 8 E.G.M. Petrakis i 0 1 a 0 1 a 1 0 a 2 1 a 3 2 b 4 3 c 5 4 c 6 5 c 7 6 d 8 7 2 b 2 1 1 2 2 3 4 5 6 3 b 3 2 2 2 2 2 4 5 6 4 c 4 3 3 3 3 2 3 4 5 5 d 5 4 4 4 4 3 3 4 4 Multimedia Information Retrieval 6 d 6 5 5 5 5 4 4 4 4 initialization cost total cost 34 Damerau-Levenstein Algorithm Basic recurrence relation (DP algorithm) edit(0,0) = 0;edit(i,0) = i; edit(0,j) = j; edit(i,j) = min{ edit(i-1,j) + 1, edit(i, j-1) + 1, edit(i-1,j-1) + d(si,ti), edit(i-2,j-2) + d(si,tj-1) + d(si-1,sj) + 1 } E.G.M Petrakis Multimedia Information Retrieval 35 Matching Algorithm Compute D(A,B) #A, #B lengths of A, B 0: null symbol R: cost of an edit operation D(0,0) = 0 for i = 0 to #A: D(i:0) = D(i-1,0) + R(A[i]0); for j = 0 to #B: D(0:j) = D(0,j-1) + R(0B[j]); for i = 0 to #A for j = 0 to #B { } E.G.M. Petrakis 1. m1 = D(i,j-1) + R(0B[j]); 2. m2 = D(i-1,j) + R(A[i] 0); 3. m3 = D(i-1,j-1) + R(A[i] B[j]); 4. D(i,j) = min{m1, m2, m3}; Multimedia Information Retrieval 36 Classical IR Models Boolean model simple based on set theory queries as Boolean expressions Vector space model queries and documents as vectors in term space Probabilistic model a probabilistic approach E.G.M Petrakis Multimedia Information Retrieval 37 Text Indexing A document is represented by a set of index terms that summarize document contents adjectives, adverbs, connectives are less useful mainly nouns (lexicon look-up) requires text pre-processing (off-line) E.G.M Petrakis Multimedia Information Retrieval 38 Indexing index Data Repository E.G.M Petrakis Multimedia Information Retrieval 39 Text Preprocessing Extract index terms from text Processing stages word separation, sentence splitting change terms to a standard form (e.g., lowercase) eliminate stop-words (e.g. and, is, the, …) reduce terms to their base form (e.g., eliminate prefixes, suffixes) construct mapping between terms and documents (indexing) E.G.M Petrakis Multimedia Information Retrieval 40 Text Preprocessing Chart from Baeza – Yates & Ribeiro – Neto, 1999 E.G.M Petrakis Multimedia Information Retrieval 41 Text Indexing Methods Main indexing methods: inverted files signature files bitmaps Size of index may exceed size of actual collection compress index to reduce storage typically 40% of size of collection E.G.M Petrakis Multimedia Information Retrieval 42 Inverted Files Dictionary: terms stored alphabetically Posting list: pointers to documents containing the term Posting info: document id, frequency of occurrence, location in document, etc. dictionary indexed by B-trees, tries or binary searched Pros: fast and easy to implement Cons: large space overhead (up to 300%) E.G.M Petrakis Multimedia Information Retrieval 43 Inverted Index index άγαλμα αγάπη … δουλειά … πρωί … ωκεανός E.G.M Petrakis posting list (1,2)(3,4) (4,3)(7,5) ……… (10,3) Multimedia Information Retrieval documents 1 2 3 4 5 6 7 8 9 10 11 44 Query Processing 1. Parse query and extract query terms 2. Dictionary lookup binary search 3. Get postings from inverted file one term at a time 4. Accumulate postings record matching document ids, frequencies, etc. compute weights 5. Rank answers by weights (e.g., frequencies) E.G.M Petrakis Multimedia Information Retrieval 45 Signature Files A filter for eliminating most of the nonqualifying documents signature: short hash-coded representation of document or query the signatures are stored and searched sequentially search returns all qualifying documents plus some false alarms the answer is searched to eliminate false alarms E.G.M Petrakis Multimedia Information 46 Retrieval Superimposed Coding Each word yields a bit pattern of size F with m bits set to 1 and the rest left as 0 size of F affects the false drop rate bit patterns are OR-ed to form the doc signature find documents having 1 in the same bits as the query E.G.M Petrakis word signature data 001 000 110 010 base 000 010 101 001 document signature 001 010 111 011 Multimedia Information Retrieval 47 Bitmaps For every term, a bit vector is stored each bit corresponds to a document set to 1 if term appears in document, 0 otherwise e.g., “word” 10100: the “word” appears in documents 1 and 3 in a collection of 5 documents Very efficient for Boolean queries retrieve bit-vectors for each query term E.G.Mcombine Petrakis Multimedia Information bit vectors with Boolean operators48 Retrieval Comparison of Text Indexing Methods Bitmaps: pros: intuitive, easy to implement and to process cons: space overhead Signature files: pros: easy to implement, compact index, no lexicon cons: many unnecessary accesses to documents Inverted files: pros: used by most search engines cons: large space overhead, index can be compressed E.G.M Petrakis Multimedia Information Retrieval 49 References “Searching Multimedia Databases by Content”, C. Faloutsos, Kluwer Academic Publishers, 1996 “Modern Information Retrieval”, R. Baeza-Yates, B. Ribeiro-Neto, Addison Wesley, 1999 “Automatic Text Processing”, Gerard Salton, Addison Wesley, 1989 Information Retrieval Links: http://www-a2k.is.tokushima-u.ac.jp/member/kita/NLP/IR.html E.G.M Petrakis Multimedia Information Retrieval 50