Context-Aware Personal Information Retrieval From Multiple Social Networks

Presented by: Sophie Le Page and Theodore Morin

Authors: Xiaogang Han, Wei Wei, Chunyan Miao, Jian-Ping Mei, and Hengjie Song

Social Network Services

• People use SNSs to collect and share previously-seen information, such as using:
  • Microblogging (Twitter)
  • Social networks (Facebook)
  • Social bookmarking (Delicious)
• Referring to and integrating previously-seen information is common
• 58-81% of web page access are re-visits, such as: 
  • ​Replying to questions on question answering websites
  • Replying to posts on SNSs

Problem

• How do we automatically retrieve the most context-relevant previously-seen web information without user intervention

For Example:

• A film lover has reviewed a movie on Facebook
• A Friend posts about the movie on Twitter
• The film lover could provide comments about the movie by retrieving the review, but may have forgotten it

Personal Web Information

• PWIs are used to specify previously-seen information on multiple SNSs
• It is challenging to make connections between the user’s context and their PWIs when the PWIs spread across multiple SNSs

Problem Statement 

• Given a session and the targeting replier, generate a query to retrieve the most relevant PWIs from the target's document collection

Problem Statement Example

Solution

Propose the Context-Aware Personal Information Retrieval (CPIR) algorithm

• Builds a query by capturing the user's information need
• Retrieves the user's most relevant PWIs

Challenges:

• Posts in the conversations are short and ambiguous
• Documents in SNSs are noisy and complex

Context-Aware Personal Information Retrieval Algorithm

Key notations:

• A Session (S) is an online conversation with:
  • An initial post p
  • A set of replies R
• Represented by the Vector Space Model
• Each term is weighted by its tf-idf score

Context-Aware Personal Information Retrieval Algorithm (continued)

1. Query formulation and expansion
2. PWIs ranking

Query Formulation and Expansion

Query Q is built by:

• Considering both replies and PWIs of all participating users
• Using PWIs of the creator and repliers to obtain richer information

Initial Post and Replies

• First, the initial p is treated as the basic query
• Next, combine the replies with p
  • Replies are weighted according to their similarities with p
• The expanded query is calculated as follows:

Methods, Techniques and External Sources

KL-divergence method

• Obtains better results than vector space based measures

Smoothing techniques

• Takes the entire vocabulary into consideration to compare two distributions 

WordNet external source

• Expands the documents before calculating similiarities

PWIs of the creator and existing repliers

• Consider PWIs of the creator and existing repliers to further exand the query
• Only the top k most relevant PWIs are selected
• The expanded query can be represented as:

PWIs Ranking

Importance Ranking

• User in the same session S share common interests (at least the topic S)
• Employ a Markov random walk model
• Rank the PWIs of a user u on implicit relationships between the web information of all users in S
• Find a subset of u's PWIs that are most relevant to the topic of the session

Final Ranking

• Use a linear combination of the two previously mentioned ranking scores:
  • Similarity between the expanded query Q and each document
  • Importance of the document in the collection of PWIs
• Obtain the final score for each document:

• The top ranked PWIs are selected as the recommendation results to the targeting replier

Data Description

• FriendFeed dataset
  • collected by monitoring the data stream on FriendFeed from 01/08/2010 to 30/09/2010 (two months)
• From these conversations select 
  • Post-reply pairs written in English
  • Repliers that have at least 50 PWIs

Manual Annotation

To construct manual annotation results:

• Randomly sample 105 post-reply pairs
  • replies are posted by 73 unique users
  • each user has ~316 PWIs
• Two volunteers manually labeled 23,046 PWIs of the repliers as relevant or irrelevant
• Tokenization and part-of speech tagging are performed to eliminate noisy terms
• Stop words are removed and terms are stemmed

Data Analysis

• 98% of conversations have at least three replies
• 78% of conversations have at least three unique repliers
• Confirms feasibility of using the conversations to model task environment to receieve past information

Data Analysis (continued)

• 65% of users use at least two services
  • Confirms documents are extracted from diverse information
• 63% of users posed more than 10 PWIs
  • Motivation to utilize PWIs of users to expand query and improve retrieval performance

Retrieval Performance

• CPIR λ=1 achieves improvement over baseline methods
  • ​Expanding the initial query with replies in the conversation enhanced context cues
  • Adding PWIs further captured the content information
• CPIR graph-based ranking algorithm further improves performance

Parameter Settings

• Optimal parameter obtained by fine tuning
• Most important parameter λ controls how to combine the ranking scores from the random walk model​

Conclusions and Future Work

Conclusion

• CPIR significantly outperforms baseline methods

Future Work

• Replace importance ranking algorithm with clustering-based techniques
• Take document recency as a factor in document ranking