Social Media Research Blog

Notes from DEBSM Workshop: Inferring Privacy Information via Social Relations

Inferring Privacy Information via Social Relations
Wanhong Xu, Xi Zhou Lei Li (presenter)

The presentation started with a neat quote: "Your social activities tell who you are". Social Networking are part of everyday life and for many of us a primary way in which we keep in touch with friends and family. Privacy is a huge problem in such networks. Li provided some statistics from a recent survey of British social network users: about  62 percent are concerned about the security of their personal data. 31% of the users falsify information to protect identity. This is a huge figure and definitely shows the level of concern.

This paper was motivated with applications for social advertising (which is a 2.2B US market). One can target users based on location, age gender etc. and social advertising allows advertisers to choose their audience. The problem comes up where users do not wish to disclose too much of their information online. The proposed solution is to automatically infer such missing information.

The authors use an undisclosed dataset for this study. The assumption is users fake personal information but not their activities. It is well known in offline-social networks that gender preference exist in friendship. however in online social networks this is not true (Jure Leskovec, WWW,2008). The key question that authors of this paper ask is : "Users may fake their personal information...But what about social activities?"

The insight that this paper offers is that certain group membership gives hints of user gender information. Joining groups has a gender preference. So the way in which one can infer the gender is to use a bipartite graph of users->groups with missing gender information using relation between users and groups. One approach is to use the User*Group matrix and build a classifier. However, they found that Naive Bayes does not work very well. Many social groups in fact dont have any gender preference. This can really hurt the classifier accuracy. So one approach they propose is to choose the discriminating social groups. i.e. groups that dominated by males or females. One major disadvantage that this technique suffers from is that the membership for users that dont join any of these groups can not be predicted. Once you restrict to the set of discriminating groups, now Naive Bayes performs well.

For users that cant be predicted they propose the use of an iterative algorithm to combine discriminative social groups and results from Naive Bayes classifier. Testing is performed by removing or making some data missing and then predicting the missing gender information.

My concern here is that the authors might have been able to avoid this disadvantage if they had used SVD to map the Groups to a set of lower dimensions that way it would automatically "cluster" the groups based on whatever the discriminating factor is (in their case the gender information). Secondly, while the authors had access to "verifiable" ground truth data, in real world how do we know the influence of fake profiles on these discriminative groups?

Notes from DEBSM Workshop: Using SentiWordNet for Multilingual Sentiment Analysis

Using SentiWordNet for Multilingual Sentiment Analysis
Kerstin Denecke

Subjectivity Analysis is about identifying a set of opinions emotions and thoughts that are expressed in text. As opposed to factual statements (objective text), the goal is to identify if there are any:

• sentiment or emotions expressed in a piece of text
• type of subjectivity can be either positive/negative opinions

The question addressed in this paper is how to analyze subjectivity and sentiments in in multilingual datasets. This was a really well motivated paper and a it was great to hear this talk especially after attending the recent tutorial by Dr. Jan Weibe at ICWSM. One motivation is that a company that sells products in many different countries might want to know the sentiments of its clients. A big difficulty in aggregating across different geographies is analyzing multilingual text.

Some of the difficulties in dealing with multilingual sentiment analysis are:

• missing language specific lexical resources
• missing linguistic resources (POS Tagger, Parser)
• missing training materials

The solution proposed by Kerstin in this paper is to rely on resources available in English and use existing training sets (from English text) to identify the subjectivity in foreign language text. This is such a simple and effective idea that my first reaction was that of surprise! I am amazed that someone had not proposed it yet!

Kerstin's approach is to take the documents and translate them to English and sentiment analysis is applied on the translated version of the document to determine the polarity. One key technology being used in this work is SentiWordNet. It is a lexical resource for opinion mining that was manually created. It offers a triple of polarity scores for each wordnet synset (pos, neg, objectivity). This is an amazing resource for anyone working on Sentiment analysis. I was unaware of SentiWordNet when we were working on the TREC Blog track 2006. In retrospect, I wish I knew of it back then.

Working with the translated text, this paper identifies each word+POS and does a lookup in sentiword to obtain the scores per synset. Next, they calculate the score for each word class (adj,noun, verbs) for a given document. Finally, the score triples are averaged across the words in the sentences. The document is given a polarity score using the triples across all the sentences.

To identify the polarity there were two approaches that were used:

• Use a rule based approach by thresholding
• Use Machine Learning (WEKA) to train a logistic regression classifier.

Evaluation was done by using IMDB archive (1000 pos,neg reviews in English) and the testset was  MPQA (English) and Amazon movie reviews (German).

Some sources of errors mentioned in the presentation:

• Statistical translation errors
• "writing errors", which I think means slang usage (scaaaaade = what a pity or aaaawful)
• The system does not consider negated structures (not bad)
• mising resolution abiguities 14 synset for "bad"

Their results indicate that ML based approach is better and obtains an accuracy of around .65.

Notes from DEBSM Workshop: Monitoring the Evolution of Interest in Blogosphere

Monitoring the evolution of interests in the blogosphere
Iraklis Varlamis (Presenter), Vasilis Vassalos, Antonis Palaios

The authors present a simple (perhaps even obvious) hypothesis that "Interests of bloggers converge around real world events".

A prototype system was built to to analyze blogs and monitor the evolution of interests. Existing search engines monitor the term popularity of words in isolation and the related terms are not clustered in any way to identify the topics. A motivating example presented was around the football world cup season, where there might be terms like "football" "round" "final" "result" "world cup" and aggregated scores for topic would be helpful rather than having to find/view the histograms for each of these terms. Also one can not know all the terms for a given topic. In many cases, multilingual content and difference in vocabulary can further make it difficult to just monitor the term popularity. Other approaches require training a topic based on a given set of training/pre-annotated document pool. This is an expensive and a semi-supervised method can be strongly effected by the selection of the initial document pool.

The goal of their paper is to build an unsupervised approach to cluster the documents and identify the evolution of interest. However, the current algorithm presented is not incremental; but they claim some recent work on an incremental version. The current version of their algorithm uses the DBScan clustering technique. Term space is clustered at the post level and their evaluation is done using inter cluster, intra cluster similarity and the utility function. They have used the BuzzMetrics dataset, however they only focus on blogs that have a post everyday, which brings the dataset down to 2500 blogs. The posts are classified into topics and then the blogs are clustered. The categories are determined by using the Cyberfiber news groups to find the topics and training documents. This is an interesting dataset that I was not aware of.

Some questions that came up were the choice of using newswire documents to learn classifiers, and also the accuracy of the classifiers. While the original hypothesis seems obvious, it can be seen as a litmus test for the efficacy of the clustering algorithms. In particular the examples presented were using the "London Bombing" Event. Any clustering algorithm that groups the related terms must at the very least pass the litmus test to show that it works.

I think that in general there are perhaps two types of events, one concerning a very specific group of individuals  ("County elections", social events, conferences etc) where online communities emerge due to offline events. The other is around real word events like "London bombings" or "Hurricane Katrina": both of which are examples of events that cut across community boundaries and which invoke reaction from almost every part of the blogosphere and almost in unison. In such cases on might see that the original community affiliation of the bloggers dont matter and tech, politics, religion and every other community might be talking about the same issue.

Notes from DEBSM Workshop: Exploiting Social Relations for Query Expansion

Exploiting Social Relations for Query Expansionand Result Ranking
Matthias Bender, Tom Crecelius, Mouna Kacimi, Sebastian Michel, Thomas Neumann, Josiane Xavier Parreira, Ralf Schenkel, Gerhard Weikumza

This paper was presented by Mouna Kacimi.

This paper is about Query expansion in social networks and addresses the question of defining the valuable content to users in rich, online communities. Mouna suggests that one way to improve the search experience is by utilizing the social network information. The paper defines a social graph and provides scoring mechanism for finding and ranking relevant content for each node. In social networks, users can be both producers and consumers of information.Typically, the user has documents (photos, posts, etc) and tags. There are relations between nodes of the same type, for example

• friendship between nodes
• similarity between text
• similarity between tags
• linkage between the documents 

Further there can be relations between nodes of different types:

• Between tags and document
• users and tags -- how imp is this tag for a user;
• user can also have a relation with documents in terms of ratings

This paper presents a generic framework for ranking relevant information for a given user; by utilizing the various elements of the social information. The scoring model can be thought of as a means of querying what your friends think about the relevance of the given information. It combines the following pieces of information:

• what your friends think of this document
• how strong is your friendship with your friend
• user rank for the friend who described the document
• document rank of the document.

Note that the authors have considered an independence assumption in computing the scores. I am not certain if this is a good assumption, but it would be one simple way of combining the scores.

In order to do query expansion, i "ask" my friends about Paris, friend's documents about paris are retrieved and other similar tags are obtained. These are used to compute the query expansion. Friendship strength is computed using the "social distance" while the tag similarity is computed based on co-occurence of the tags.

I was surprised by their finding that social search works better in flickr but not very well in del.icio.us. I think that this is because the social networks of del.icio.us are very sparse. The meaning of friendship is also different here (aka friends and fans).

However, Mouna also pointed out that an improved version of their technique has been accepted at SIGIR 2008 and has a more detailed analysis. I look forward to checking out their extended paper. What I really liked about this paper was that it was a generic framework for combining social information (network, tags, documents, etc) and can be extended in ways that would be suitable for a particular type of network or task.

 

ICDE Workshop on Data Engineering for Blogs Social Media and Web 2.0

The Data Engineering for Blogs, Social Media and Web 2.0 workshop was held in conjunction with ICDE 2008. There were some really interesting papers and the invited talk was given by  Prof. Ravi Sandhu. I had the opportunity to present some of our work at this workshop. My talk was about "Communities in Social media" which was about some of our recent work on "Approximating the Community Structure of the Long Tail" (ICWSM) and an overview of our previous and recent research in this area.

As Matt noticed already, not all the authors have their papers online, yet. I am still trying to figure out what the IEEE conference policy is for uploading the PDFs of the papers on the workshop website. Perhaps, I shall try to request the authors to upload the copies and link to them from the workshop website. In the following series of posts, I will share some of my notes from the different talks.

And oh yesss! the fact that the conference was in Cancun was just great! :-D