1. Web-Content Mining -Akanksha Dombe

2. Specifies  The WWW is huge, widely distributed, global information service centre for  Information services: news, advertisements, consumer information, financial management, education, government, e-commerce, etc.  Hyper-link information  Access and usage information  WWW provides rich sources of data for data mining

3. The Web: Opportunities & Challenges 1.The amount of information on the Web is huge 2.The coverage of Web information is very wide and diverse 3.Information/data of almost all types exist on the Web 4.Much of the Web information is semi-structured 5.Much of the Web information is linked 6.Much of the Web information is redundant

4. The Web: Opportunities & Challenges 7.The Web is noisy 8.The Web is also about services 9.The Web is dynamic 10.Above all, the Web is a virtual society 11.The Web consists of surface Web and deep Web.  Surface Web: pages that can be browsed using a browser.  Deep Web: databases that can only be accessed through parameterized query interfaces

5. What is Web Data ? What is Web Data ?  Web data is 1.Web content –text,image,records,etc. 2.Web structure –hyperlinks,tags,etc. 3.Web usage –http logs,app server logs,etc. 4.Intra-page structures 5.Inter-page structures 6.Supplemental data 1.Profiles 2.Registration information 3.Cookies

7. Web Mining  Web Mining is the use of the data mining techniques to automatically discover and extract information from web documents/services  Web mining is the application of data mining techniques to find interesting and potentially useful knowledge from web data  Web mining is the application of data mining techniques to extract knowledge from web data, including web documents, hyperlinks between documents, usage logs of web sites, etc.

8. Web Mining Web Mining is the use of the data mining techniques to automatically discover and extract information from web documents/services Discovering useful information from the World-Wide Web and its usage patterns My Definition: Using data mining techniques to make the web more useful and more profitable (for some) and to increase the efficiency of our interaction with the web

9. Why Mine the Web?  Enormous wealth of information on Web  Financial information (e.g. stock quotes)  Book/CD/Video stores (e.g. Amazon)  Restaurant information  Car prices  Lots of data on user access patterns  Web logs contain sequence of URLs accessed by users  Possible to mine interesting nuggets of information  People who ski also travel frequently to Europe  Tech stocks have corrections in the summer and rally from November until February

10.  The Web is a huge collection of documents except for  Hyper-link information  Access and usage information  The Web is very dynamic  New pages are constantly being generated  Challenge: Develop new Web mining algorithms and adapt traditional data mining algorithms to  Exploit hyper-links and access patterns  Be incremental Why is Web Mining Different?

11. Web Mining: Subtasks Web Mining: Subtasks  Resource finding  Retrieving intended documents  Information selection/pre-processing  Select and pre-process specific information from selected documents  Generalization  Discover general patterns within and across web sites  Analysis  Validation and/or interpretation of mined patterns

12. Web Mining Issues  Size  Grows at about 1 million pages a day  Google indexes 9 billion documents  Number of web sites  Netcraft survey says 72 million sites  (http://news.netcraft.com/archives/web_server_survey.html)http://news.netcraft.com/archives/web_server_survey.html  Diverse types of data  Images  Text  Audio/video  XML  HTML

13.  E-commerce (Infrastructure)  Generate user profiles  Targetted advertizing  Fraud  Similar image retrieval  Information retrieval (Search) on the Web  Automated generation of topic hierarchies  Web knowledge bases  Extraction of schema for XML documents  Network Management  Performance management  Fault management Web Mining Applications

14. Web Mining Taxonomy

15. Web Data Mining  Use of data mining techniques to automatically discover interesting and potentially useful information from Web documents and services.  Web mining may be divided into three categories: 1. Web content mining 2. Web structure mining 3. Web usage mining

16. What is “Web Content mining?”

17. Web Content Mining  Discovery of useful information from web contents / data / documents  Web data contents: 1. text, 2. image, 3.audio, 4.video, 5.metadata and 6.hyperlinks

18. Web Content Mining  Examine the contents of web pages as well as result of web searching  Can be thought of as extending the work performed by basic search engines  Search engines have crawlers to search the web and gather information, indexing techniques to store the information, and query processing support to provide information to the users  Web Content Mining is: the process of extracting knowledge from web contents

19. Web Content Mining  It provides no information about structure of content that we are searching for and no information about various categories of documents that are found.  Need more sophisticated tools for searching or discovering Web content.

20. Web Content mining  Discovering useful information from contents of Web pages.  Web content is very rich consisting of textual, image, audio, video etc and metadata as well as hyperlinks.  The data may be unstructured (free text) or structured (data from a database) or semi-structured (html) although much of the Web is unstructured.

21. Web Content Data Structure  Unstructured – free text  Semi-structured – HTML  More structured – Table or Database generated HTML pages  Multimedia data – receive less attention than text or hypertext

22. Web Content mining  Web content mining is related to data mining and text mining  It is related to data mining because many data mining techniques can be applied in Web content mining.  It is related to text mining because much of the web contents are texts.  Web data are mainly semi-structured and/or unstructured, while data mining is structured and text is unstructured.

23. Web Content Data Structure  Web content consists of several types of data  Text, image, audio, video, hyperlinks.  Unstructured – free text  Semi-structured – HTML  More structured – Data in the tables or database generated HTML pages  Note: much of the Web content data is unstructured text data.

24. Semi-structured Data  Content is, in general, semi-structured  Example:  Title  Author  Publication_Date  Length  Category  Abstract  Content

25. Web Content Mining: IR View  Unstructured Documents  Bag of words, or phrase-based feature representation  Features can be boolean or frequency based  Features can be reduced using different feature selection techniques  Word stemming, combining morphological variations into one feature

26. Web Content Mining: IR View  Semi-Structured Documents  Uses richer representations for features, based on information from the document structure (typically HTML and hyperlinks)  Uses common data mining methods (whereas unstructured might use more text mining methods)

27. Web Content Mining: DB View  Tries to infer the structure of a Web site or transform a Web site to become a database  Better information management  Better querying on the Web  Can be achieved by:  Finding the schema of Web documents  Building a Web warehouse  Building a Web knowledge base  Building a virtual database

28. Web Content Mining: DB View  Mainly uses the Object Exchange Model (OEM)  Represents semi-structured data (some structure, no rigid schema) by a labeled graph  Process typically starts with manual selection of Web sites for content mining  Main application: building a structural summary of semi-structured data (schema extraction or discovery)

29. Tech for Web Content Mining  Classifications  Clustering  Association

30. Web Content Mining : Topics  Structured data extraction  Unstructured text extraction  Sentiment classification, analysis and summarization of consumer reviews  Information integration and schema matching  Knowledge synthesis  Template detection and page segmentation

31. Structured Data Extraction  Most widely studied research topic  A large amount of information on the Web is contained in regularly structured data objects (retrieved from databases)Such Web data records are important they often present the essential information of their host pages, e.g., lists of products and services

32. Structured Data Extraction  Applications: integrated and value-added services, e.g., Comparative shopping, meta-search & query, etc

33. Structured Data Extraction :Approaches 1.Wrapper Generation 2.Wrapper Induction or Wrapper Learning 3.Automatic Approach

34. Structured Data Extraction :Approaches  Wrapper Generation Write an extraction program for each website based on observed format patterns  Labor intensive & time consuming

35. 35

36. 36

37. CS511, Bing Liu, UIC 37

38.  Automatic Approach  Structured data objects on the web are normally database records  Retrieved from databases & displayed in web pages with fixed templates  Find patterns / grammars from the web pages & then use them to extract data  e. g. IEPAD, MDR, ROADRUNNER, EXALG etc 38

39.  Wrapper Induction or Wrapper Learning  Main technique currently  The user first manually labels a set of trained pages  A learning system then generates rules from the training pages  The resulting rules are then applied to extract target items from web pages  e.g. WIEN, Stalker, BWI, WL etc 39

40.  Supervised Learning  Supervised learning is a ‘machine learning’ technique for creating a function from training data.  Documents are categorized  The output can predict a class label of the input object (called classification).  Techniques used are  Nearest Neighbor Classifier  Feature Selection  Decision Tree

41.  Removes terms in the training documents which are statistically uncorrelated with the class labels  Simple heuristics  Stop words like “a”, “an”, “the” etc.  Empirically chosen thresholds for ignoring “too frequent” or “too rare” terms  Discard “too frequent” and “too rare terms”

42. Examples of Discovered Patterns  Association rules  98% of AOL users also have E-trade accounts  Classification  People with age less than 40 and salary > 40k trade on- line  Clustering  Users A and B access similar URLs  Outlier Detection  User A spends more than twice the average amount of time surfing on the Web

43.  Important for improving customization  Provide users with pages, advertisements of interest  Example profiles: on-line trader, on-line shopper  Generate user profiles based on their access patterns  Cluster users based on frequently accessed URLs  Use classifier to generate a profile for each cluster  Engage technologies  Tracks web traffic to create anonymous user profiles of Web surfers  Has profiles for more than 35 million anonymous users

44.  Ads are a major source of revenue for Web portals (e.g., Yahoo, Lycos) and E-commerce sites  Plenty of startups doing internet advertizing  Doubleclick, AdForce, Flycast, AdKnowledge  Internet advertizing is probably the “hottest” web mining application today

45.  Scheme 1:  Manually associate a set of ads with each user profile  For each user, display an ad from the set based on profile  Scheme 2:  Automate association between ads and users  Use ad click information to cluster users (each user is associated with a set of ads that he/she clicked on)  For each cluster, find ads that occur most frequently in the cluster and these become the ads for the set of users in the cluster

46.  Use collaborative filtering (e.g. Likeminds, Firefly)  Each user Ui has a rating for a subset of ads (based on click information, time spent, items bought etc.)  Rij - rating of user Ui for ad Aj  Problem: Compute user Ui’s rating for an unrated ad Aj A1A2A3 ? Internet Advertizing

47.  Key Idea: User Ui’s rating for ad Aj is set to Rkj, where Uk is the user whose rating of ads is most similar to Ui’s  User Ui’s rating for an ad Aj that has not been previously displayed to Ui is computed as follows:  Consider a user Uk who has rated ad Aj  Compute Dik, the distance between Ui and Uk’s ratings on common ads  Ui’s rating for ad Aj = Rkj (Uk is user with smallest Dik)  Display to Ui ad Aj with highest computed rating Internet Advertizing

48.  With the growing popularity of E-commerce, systems to detect and prevent fraud on the Web become important  Maintain a signature for each user based on buying patterns on the Web (e.g., amount spent, categories of items bought)  If buying pattern changes significantly, then signal fraud  HNC software uses domain knowledge and neural networks for credit card fraud detection

49.  Given:  A set of images  Find:  All images similar to a given image  All pairs of similar images  Sample applications:  Medical diagnosis  Weather predication  Web search engine for images  E-commerce

50.  QBIC, Virage, Photobook  Compute feature signature for each image  QBIC uses color histograms  WBIIS, WALRUS use wavelets  Use spatial index to retrieve database image whose signature is closest to the query’s signature  WALRUS decomposes an image into regions  A single signature is stored for each region  Two images are considered to be similar if they have enough similar region pairs

51. Query image

52.  Today’s search engines are plagued by problems:  the abundance problem (99% of info of no interest to 99% of people)  limited coverage of the Web (internet sources hidden behind search interfaces)  Largest crawlers cover < 18% of all web pages  limited query interface based on keyword- oriented search  limited customization to individual users

53.  Today’s search engines are plagued by problems:  Web is highly dynamic  Lot of pages added, removed, and updated every day  Very high dimensionality

54.  Use Web directories (or topic hierarchies)  Provide a hierarchical classification of documents (e.g., Yahoo!)  Searches performed in the context of a topic restricts the search to only a subset of web pages related to the topic RecreationScienceBusinessNews Yahoo home page SportsTravelCompaniesFinanceJobs

55.  In the Clever project, hyper-links between Web pages are taken into account when categorizing them  Use a bayesian classifier  Exploit knowledge of the classes of immediate neighbors of document to be classified  Show that simply taking text from neighbors and using standard document classifiers to classify page does not work  Inktomi’s Directory Engine uses “Concept Induction” to automatically categorize millions of documents

56.  Objective: To deliver content to users quickly and reliably Traffic management Fault management Service Provider Network Router Server

57.  While annual bandwidth demand is increasing ten- fold on average, annual bandwidth supply is rising only by a factor of three  Result is frequent congestion at servers and on network links  during a major event (e.g., princess diana’s death), an overwhelming number of user requests can result in millions of redundant copies of data flowing back and forth across the world  Olympic sites during the games  NASA sites close to launch and landing of shuttles

58.  Key Ideas  Dynamically replicate/cache content at multiple sites within the network and closer to the user  Multiple paths between any pair of sites  Route user requests to server closest to the user or least loaded server  Use path with least congested network links  Akamai, Inktomi

59. Service Provider Network Router Server Request Congested server Congested link

60.  Need to mine network and Web traffic to determine  What content to replicate?  Which servers should store replicas?  Which server to route a user request?  What path to use to route packets?  Network Design issues  Where to place servers?  Where to place routers?  Which routers should be connected by links?  One can use association rules, sequential pattern mining algorithms to cache/prefetch replicas at server

61.  Fault management involves  Quickly identifying failed/congested servers and links in network  Re-routing user requests and packets to avoid congested/down servers and links  Need to analyze alarm and traffic data to carry out root cause analysis of faults  Bayesian classifiers can be used to predict the root cause given a set of alarms

62. Total Sites Across All Domains August 1995 - October 2007

63.  Web data sets can be very large  Tens to hundreds of terabytes  Cannot mine on a single server!  Need large farms of servers  How to organize hardware/software to mine multi-terabye data sets  Without breaking the bank!

64.  Structured Data  Unstructured Data  OLE DB offers some solutions!

65.  Pages contain information  Links are ‘roads’  How do people navigate the Internet   Web Usage Mining (clickstream analysis)  Information on navigation paths available in log files  Logs can be mined from a client or a server perspective

66.  Why analyze Website usage?  Knowledge about how visitors use Website could  Provide guidelines to web site reorganization; Help prevent disorientation  Help designers place important information where the visitors look for it  Pre-fetching and caching web pages  Provide adaptive Website (Personalization)  Questions which could be answered  What are the differences in usage and access patterns among users?  What user behaviors change over time?  How usage patterns change with quality of service (slow/fast)?  What is the distribution of network traffic over time?

69.  Analog – Web Log File Analyser  Gives basic statistics such as  number of hits  average hits per time period  what are the popular pages in your site  who is visiting your site  what keywords are users searching for to get to you  what is being downloaded  http://www.analog.cx/

73.  Content is, in general, semi-structured  Example:  Title  Author  Publication_Date  Length  Category  Abstract  Content

74.  Many methods designed to analyze structured data  If we can represent documents by a set of attributes we will be able to use existing data mining methods  How to represent a document?  Vector based representation (referred to as “bag of words” as it is invariant to permutations)  Use statistics to add a numerical dimension to unstructured text

75.  A document representation aims to capture what the document is about  One possible approach:  Each entry describes a document  Attribute describe whether or not a term appears in the document

76.  Another approach:  Each entry describes a document  Attributes represent the frequency in which a term appears in the document

77.  Stop Word removal: Many words are not informative and thus  Irrelevant for document representation the, and, a, an, is, of, that, …  Stemming: reducing words to their root form (Reduce dimensionality)  A document may contain several occurrences of words like fish, fishes, fisher, and fishers. But would not be retrieved by a query with the keyword “fishing”  Different words share the same word stem and should be represented with its stem, instead of the actual word “Fish”