J. Bengel, S. Gauch, E. Mittur, R. Vijayaraghavan - PowerPoint PPT Presentation

Title: J. Bengel, S. Gauch, E. Mittur, R. Vijayaraghavan


1
Chat Room Topic Detection Using Classification

• J. Bengel, S. Gauch, E. Mittur,R.
  Vijayaraghavan
• Presented by Dr. Susan Gauch

2
Outline

• Motivation, Goals
• Related Work
• ChatTrack Architecture
• Scenarios
• Conclusion
• Future Work

3
Motivation

• Popularity of Chat Rooms IM soaring for both
  adults teenagers
• Consumers entertainment value, meet new people
  globally
• Corporations business contacts, virtual
  meetings
• 2005 IM will surpass e-mail for primary
  electronic communications (Gartner Group)
• Difficult to retain track conversations
• Criminals embedded in chat culture remain hidden
• Disguise identity Lure children
• Commit corporate/homeland espionage

4
Motivation (2)

• Chat provides a new tool for collaboration
  between criminals
• Recruit new terrorists and sympathizers
• Dissemination of criminal plots
• Detection of illegal activities may enhance crime
  prevention
• Human monitors could monitor chat discussions
• Invasive (privacy concerns)
• Human error
• Expensive!

5
Goals

• Assist in crime detection by producing a
  concept-based profile, summarizing chat data
• over entire public chat sessions
• over individuals 11 chat
• to analyze silent listeners interests
• Chat Profile allows focus on domains of criminal
  interest, national security, protection of
  children
• Examples
• science/microbiology (anthrax)
• music/instruments/violins (Stradivarius heist)

6
Related Work

• Topic Identification longstanding problem
• Chat structured differently than normal writing
• Butterfly Samples 30-seconds of chat data to
  determine chat room topics search for topics of
  interest Van Dyke, et. al.
• Wayne, C. L. text classification for
  discovering, threading, retrieving from data
  streams
• Bingham, E., et. al. topic ID (from news
  sources using complexity pursuit

7
Related Work (2)

• PeopleGarden creates data portraits that
  represent interactions between users Xiong, R.,
  et. al.
• IBM WebFountain assembles bodies of knowledge
  from various sources (web, chat rooms, email)
• Elnahrawy, E. evaluated several text
  classification techniques of chat data on four
  categories
• Our work complements these systems
• combination ? powerful analysis

8
Architecture
Client
Internet
Chat DataArchive (XML/SQL)
ConceptDatabase
IRCClient
IRCClient
Chat Server(with ChatLog)
Classifier
Indexer
Administrator /Intelligence Agent
ChatProfile
ChatRetrieve
9
Chat Archival

• IRC Bot and IRC Server augmented w/ ChatLog
  Library

• Bot collects public chat data
• Server collects both public/private data
• All activity recorded (e.g., joins, parts,
  messages, nickname changes, etc.)

10
Chat Archival (2)
2004-04-17
085750 jmb jayhawk
2004-04-17
085814 Jason
jayhawk
There is going to be weather, whether or
not. Uh oh, Ill be RIGHT back!
2004-04-17
085819 jmb 2004-04-17
085946 Aliceer jayhawk
Poof..Left in a hurry! Must be a tornado
outside his door or something. lol

• XML chat data ? SQL database

• ChatLog Library XML schema can be used for
  almost any client/server-based system

11
Chat Profile

• Text classification to create profile of chat
  data
• Classification based on Vector Space model
• Training
• Use pre-defined concepts example text -- from
  ODP
• Each category Vector of representative keywords
  weights (tf idf )
• Trained on 1,565 concepts from top 3 levels of
  ODP
• Selection based on empirical studies
• Additional self-trained categories can be supplied

12
Chat Profile (2)

• User profile focuses on one chat participant
• Session profile filters by chat room name only
• Analyst selects criteria of interest

13
Chat Profile (3)

• Chat data collected from archive (Stop word
  removal Porter stemming)
• Classification performed once chat utterances
  collected
• Classifier creates vector of keywords from chat
  data
• Similarity measure
• Determines similarity between chat data vector
  vectors for each trained concept
• Concepts sorted top matches returned
• Represented in hierarchical fashion
• Asterisks represent relative concept weights at
  each level

14
Chat Profile (4)
Classification for hacking
15
Chat Profile (5)

• Intelligence agencies interests constantly
  evolve
• New world events
• Threats
• Updated intelligence
• ChatProfile allows new concepts to be added for
  profile recognition
• Agent would make use of concept database tuned
  for categories of interest only
• Omit uninteresting concepts (idle chat)

16
Chat Profile (6)
Classification forhacking(Augmented Training)
17
Classification in american-politics(two hours
in January 2004 Undernet) SESSION PROFILE
18
Classification in american-politics(two hours
in January 2004 Undernet) USER PROFILE
selected one public chat participant
19
Chat Retrieve

• Some profiles warrant further analysis
• Agent/admin needs ability to recall chat
  session linked to profiles in question
• Indexing system keeps data current
• Traditional indexing systems index from
  scratch every time new data appears

ChatRetrieve provides session details
20
Chat Retrieve (2)
IRCClient
Chat DataArchive (XML/SQL)
Chat Participants
Internet
Chat Server(ChatLog)
IRCClient
IncrementalIndexer
Inverted Index
Web-Based Retrieval System
21
Chat Retrieve (3)

• Queries based on
• speaker name
• keywords
• date/time range
• (combo of above)
• Keyword retrieval based on tf idf

22
Chat Retrieve (4)

• Selecting chat room name replays chat session
• Includes utterances spoken by all participants
• Tracks all chat room participants
• Even if they do not contribute to sessions
• Ability to search by listener

23
Conclusions
ChatTrack Technology is a framework that
enhances todays client and server-based chat
systems.

• ChatTrack Provides agents with new tools for
  vigilance against crime
• ChatProfile generates conceptual profiles from
  chat data
• Filter based on listener/sender, date/time, chat
  room session name
• ChatRetrieve facilitates manual analysis for
  session retrieval by agents, administrators,
  parents
• Reduces manual efforts to classify chat sessions

24
Future

• Automate temporal analysis of chat room topics
  (ChatTrend)
• Sudden change of interests may indicate odd
  behavior
• e.g., an upswing in discussions about public
  water utilities
• Visualize user profiles to inspect interests,
  topic shifts

25
Future (ChatTrend)

• CHATTREND SCREENSHOT HERE

26
Future (ChatTrend 2)

• CHATTREND SCREENSHOT 2 HERE

27
Future (2)

• Preprocess seemingly meaningless chatter into
  meaningful words
• brb be right back
• lol laughing out loud
• heehehehee laughing
• Chat utterance clustering identify similar chat
  content (thesis project)
• Provide ability to retrieve, replay sessions from
  archive based on topics
• Keyword search nitroglycerin or pipe bomb
• Topic search bomb making

28
Future (3)

• Proactively monitor children
• Restrict types of messages children send/receive
• Corporate monitoring
• Archive virtual meetings corporate liability
• Chat Networks
• Identify rogue users / Spambots
• Analyze metrical habits, rhetorical devices,
  polysyllabic words
• Language modeling identify user fingerprint
  (such as in analysis of Shakespeare literature)
• Detect on-line predators
• Investigate users having both adult and
  child-like language models

29
More Information

• website has many
  components available for download
• XML Chat Log schema
• ChatLog Library (in C)
• Modified IRC chat server
• ChatTrackBot (IRC data collection)

http//www.ittc.ku.edu/chattrack/