Information Retrieval CSCI

Transcription

1 Information Retrieval CSCI My name is Anwar Alhenshiri My is: I prefer: The course website is: 5/6/2012 1

2 Introductory Lecture Information Retrieval 5/6/2012 2

3 Q6. When you gather information from the Web(say for a project, a school report you are doing, or a trip you are planning to take), how many times do you usually change your query to get relevant results? A- Never, I always get what I want in the first attempt B- One time C- Two Times D- Three Times E- More 5/6/2012 3

4 IR and Databases Data Attributes Queries Results IR Databases 5/6/2012 4

5 IR and Databases IR Databases Data unstructured structured Attributes Queries Results 5/6/2012 5

6 IR and Databases IR Databases Data unstructured structured Attributes vague well defined Queries Results 5/6/2012 6

7 IR and Databases IR Databases Data unstructured structured Attributes vague well defined Queries Results keyword and features SQL defined 5/6/2012 7

8 IR and Databases IR Databases Data unstructured structured Attributes vague well defined Queries keyword and features Results imprecise exact SQL defined 5/6/2012 8

9 Definition Information retrieval (IR) is: finding material (usually documents). of an unstructured nature (usually text) that satisfies an information need from within large collections (usually stored on computers). (Manning et al., 2009) 5/6/2012 9

10 Docs DB Introduction Index Terms Doc abstract match Information Need Query 5/6/

11 Definitions A database is a collection of documents. A document is a sequence of terms, expressing ideas about some topic in a natural language. A term is a semantic unit, a word, phrase, or potentially root of a word. A query is a request for documents pertaining to some topic. 5/6/

12 So, what would be the job of an information retrieval system? 5/6/

13 Information Retrieval An Information Retrieval (IR) System attempts to find relevant documents to respond to a user s request. The real problem boils down to matching the language of the query to the language of the document 5/6/

14 IR Problems Simply matching on words is a very brittle approach. One word can have a zillion different semantic meanings Consider the word Take take a place at the table take money to the bank take a picture take a lot of time take drugs 5/6/

15 IR Problems, cont d You can t even tell what part of speech a word has: I saw her duck A query that searches for pictures of a duck will find documents that contain I saw her duck away from the ball falling from the sky 5/6/

16 IR Problems, cont d Proper Nouns often use regular old nouns Consider a document with a man named Abraham owned a Lincoln A word matching query for Abraham Lincoln may well find the above document. 5/6/

17 What is Different about IR from the Rest of Computer Science Most algorithms in computer science have a right answer: Consider the two problems: Sort the following ten integers Find the highest integer Now consider: Find the document most relevant to hippos in the zoo 5/6/

18 Measuring Effectiveness heuristic vs. exact An algorithm is deemed incorrect if it does not have a right answer. A heuristic tries to guess something close to the right answer. Heuristics are measured on how close they come to a right answer. IR techniques are essentially heuristics because we do not know the right answer. So we have to measure how close to the right answer we can come. 5/6/

19 Information Retrieval 5/6/

20 Back to the Definition of IR 5/6/

21 Unstructured (text) vs. Structured (database) data in 1996 Market 5/6/2012 capitalization/capitalisation (often market cap) is a measurement of size of a business enterprise (corporation) equal to the share price times the number of shares outstanding of a public company. 21

22 Unstructured (text) vs. Structured (database) data in /6/

23 Unstructured Data Typically refers to free text Allows Keyword queries including operators More sophisticated concept queries e.g., find all web pages dealing with drug abuse Classic model for searching text documents 5/6/

24 Semi-Structured Data In fact almost no data is unstructured E.g., this slide has distinctly identified zones such as the Title and Bullets Facilitates semi-structured search such as Title contains data AND Bullets contain search to say nothing of linguistic structure 5/6/

25 More Sophisticated Semi-Structured Search Title is about Object Oriented Programming AND Author something like stro*rup where * is the wild-card operator Issues: how do you process about? how do you rank results? The focus of XML search (IIR chapter 10) 5/6/

26 The Web and its Challenges Unusual and diverse documents Unusual and diverse users, queries, information needs Beyond terms, exploit ideas from social networks link analysis, clickstreams... How do search engines work? And how can we make them better? 5/6/

27 More Sophisticated Information Retrieval Cross-language information retrieval Question answering Summarization Text mining 5/6/

28 Sec. 1.1 Basic assumptions of Information Retrieval Collection: Fixed set of documents Goal: Retrieve documents with information that is relevant to the user s information need and helps the user complete a task 5/6/

29 Again from the Definition Document is Set of terms Bag of terms Sequence of terms Each choice has consequences term is used instead of word to signal more general possibilities: serial numbers, nonsense, etc. 5/6/

30 Modeling: query Query Basic query is one term Multi-term query is List of terms OR model: some terms AND model: all terms Boolean combination of terms Other constraints? 5/6/

31 Information Retrieval (NOTION) User wants information from a collection of objects : information need User formulates need as a query Language of information retrieval system System finds objects that satisfy query System presents objects to user in useful form User determines which objects from among those presented are relevant 5/6/

32 Information Retrieval (NOTION), cont d Define each of the words in quotes Information object Query Satisfying objects Useful presentation Notion of relevance is critical What really want? Insufficient structure for exact retrieval Develop algorithms for the search and retrieval tasks 5/6/

33 Documents Early digital searches digital card catalog: subject classifications, keywords Full text : words + English structure No meta-structure Classic study Gerald Salton SMART project 1960 s 5/6/

34 Scaling What are attributes changing from 1960 s to online searches of today? Some answers: Much much larger collections Heterogeneous collections Collections dynamic: documents come, go, change Decentralized / distributed collections More diverse users Use for relevance? More demanding users More complex queries Much much more computing resources 5/6/

35 Scaling, cont d How do these changes change problem? Some thoughts: lower concentration of clues i.e. important words computing power through clustering more complex algorithms others? 5/6/

36 The Classic Search Model TASK Info Need Verbal form Query Misconception? Mistranslation? Misformulation? Get rid of mice in a politically correct way Info about removing mice without killing them How do I trap mice alive? mouse trap SEARCH ENGINE Query Results Corpus 5/6/2012 Refinement 36

37 Readings Please read Chapter 1 of the Modern Information Retrieval Book. Credits: 5/6/

38 Information Retrieval Zipf Law 5/6/

39 Zipf Law Zipf's law states that given some corpus of natural language utterances, the frequency of any word is inversely proportional to its rank in the frequency table. Meaning that: Too frequent terms are less important. Too rare terms are less important!? 5/6/

40 A log-log plot of Words in Wikipedia A plot of word frequency in Wikipedia (November 27, 2006). The plot is in log-log coordinates. x is rank of a word in the frequency table; y is the total number of the word s occurrences. Most popular words are the, of and and, as expected. Zipf's law corresponds to the upper linear portion of the curve, roughly following the green (1/x) line. The Zipf constant of an English corpus is close to /6/

41 Zipf Law, cont d Zipf's law is most easily observed by plotting the data on a loglog graph. The axes being X= log(rank order) and Y=log(frequency). For example, the word "the" would appear at x = log(1), y = log(69971) is the number of times the word the appears in a hypothetical corpus. 1 is the rank of the word the. The data conform to Zipf's law to the extent that the plot is almost linear. 5/6/

42 Zipf Law, cont d To apply Zipf s law: Terms in the corpus of interest may be organized so that: One column carries the ranks of the terms starting with the most common terms with rank 1. Continue increasing the number in the first column as more terms appear. Ties do not matter. The second column holds the number of times each term appears in the corpus. 5/6/

43 Zipf Law Formally, cont d Zipf Law, what we can determine: R * P = A P is the probability of the number of occurrences of the term = n N A is the Zipf value of the term. R n n N = A R n = AN n The number of terms that occur n times: I n = R n R n+1 I n = AN n AN n+1 = AN n n+1 I n AN = 1 n n+1 When the frequency is 1, we get the highest rank: In /An = 1 / n (n+1) = how many terms occur n times. 5/6/

44 Example n (Rank) 1 Percentage n(n + 1) % of terms occur ONCE Around 66% of terms occur at most TWICE Around 75% of terms occur at most three times 4 5/6/

45 Zipf Law Allows us to decide on what terms to keep and what terms to ignore when indexing. If Zipf law holds, the loglog plot will be a straight line with a slope close to -1. See figure. 5/6/

46 Summary The simplest case of Zipf's law is a "1/f function". Given a set of Zipfian distributed frequencies, sorted from most common to least common, the second most common frequency will occur ½ as often as the first. The third most common frequency will occur 1/3 as often as the first. The n th most common frequency will occur 1/n as often as the first. However, this cannot hold exactly, because items must occur an integer number of times: there cannot be 2.5 occurrences of a word. Nevertheless, over fairly wide ranges, and to a fairly good approximation, many natural phenomena obey Zipf's law. 5/6/