Maarten Clements

Personalization of tagging systems

Social media systems have encouraged end user participation in the Internet, for the purpose of storing and distributing Internet content, sharing opinions and maintaining relationships. Collaborative tagging allows users to annotate the resulting user-generated content, and enables effective retrieval of otherwise uncategorised data. However, compared to professional web content production, collaborative tagging systems face the challenge that end-users assign tags in an uncontrolled manner, resulting in unsystematic and inconsistent metadata. This paper introduces a framework for the personalization of social media systems. We pinpoint three tasks that would benefit from personalization: collaborative tagging, collaborative browsing and collaborative search. We propose a ranking model for each task that integrates the individual users tagging history in the recommendation of tags and content, to align its suggestions to the individual user preferences. We demonstrate on two real data sets that for all three tasks, the personalized ranking should take into account both the users own preference and the opinion of others.

The task-dependent effect of tags and ratings on social media access

Recently, online social networks have emerged that allow people to share their multimedia files, retrieve interesting content, and discover like-minded people. These systems often provide the possibility to annotate the content with tags and ratings. Using a random walk through the social annotation graph, we have combined these annotations into a retrieval model that effectively balances the personal preferences and opinions of like-minded users into a single relevance ranking for either content, tags, or people. We use this model to identify the influence of different annotation methods and system design aspects on common ranking tasks in social content systems. Our results show that a combination of rating and tagging information can improve tasks like search and recommendation. The optimal influence of both sources on the ranking is highly dependent on the retrieval task and system design. Results on content search and tag suggestion indicate that the profile created by a users annotations can be used effectively to adapt the ranking to personal preferences. The random walk reduces sparsity problems by smoothly integrating indirectly related concepts in the relevance ranking, which is especially valuable for cold-start users or individual tagging systems like YouTube and Flickr.

The influence of personalization on tag query length in social media search

Social content systems contain enormous collections of unstructured user-generated content, annotated by the collaborative effort of regular Internet users. Tag-clouds have become popular interfaces that allow users to query the database of these systems by clicking relevant terms. However, these single click queries are often not expressive enough to effectively retrieve the desired content. Users have to use multiple clicks or type longer queries to satisfy their information need. To enhance the predicted content ranking we use a random walk model that effectively integrates the users preference and semantically related query terms. We use the collaborative annotations from a popular on-line book catalog to create a social annotation graph and study the effect of personalization and smoothing for increasing query lengths. We show that personalization and smoothing allow the user to find equally relevant content with fewer query terms compared to a frequency based content ranking with TF-IDF weighing. As expected, we see that the influence of the random walk model disappears if users type more detailed queries. Finally, we discuss the observations with respect to synonyms and homographs which are well known to hamper the performance of information retrieval systems.

Detecting synonyms in social tagging systems to improve content retrieval

Collaborative tagging used in online social content systems is naturally characterized by many synonyms, causing low precision retrieval. We propose a mechanism based on user preference profiles to identify synonyms that can be used to retrieve more relevant documents by expanding the users query. Using a popular online book catalog we discuss the effectiveness of our method over usual similarity based expansion methods.

Integration of Known Transcription Factor Binding Site Information and Gene Expression Data to Advance from Co-Expression to Co-Regulation

The common approach to find co-regulated genes is to cluster genes based on gene expression. However, due to the limited information present in any dataset, genes in the same cluster might be co-expressed but not necessarily co-regulated. In this paper, we propose to integrate known transcription factor binding site information and gene expression data into a single clustering scheme. This scheme will find clusters of co-regulated genes that are not only expressed similarly under the measured conditions, but also share a regulatory structure that may explain their common regulation. We demonstrate the utility of this approach on a microarray dataset of yeast grown under different nutrient and oxygen limitations. Our integrated clustering method not only unravels many regulatory modules that are consistent with current biological knowledge, but also provides a more profound understanding of the underlying process. The added value of our approach, compared with the clustering solely based on gene expression, is its ability to uncover clusters of genes that are involved in more specific biological processes and are evidently regulated by a set of transcription factors.

Finding wormholes with flickr geotags

We propose a kernel convolution method to predict similar locations (wormholes) based on human travel behaviour. A scaling parameter can be used to define a set of relevant users to the target location and we show how the geotags of these users can effectively be aggregated to predict a ranking of similar locations. We evaluate results on world and city level using several independent test collections.

Exploiting Positive and Negative Graded Relevance Assessments for Content Recommendation

Social media allow users to give their opinion about the available content by assigning a rating. Collaborative filtering approaches to predict recommendations based on these graded relevance assessments are hampered by the sparseness of the data. This sparseness problem can be overcome with graph-based models, but current methods are not able to deal with negative relevance assessments. We propose a new graph-based model that exploits both positive and negative preference data. Hereto, we combine in a single content ranking the results from two graphs, one based on positive and the other based on negative preference information. The resulting ranking contains less false positives than a ranking based on positive information alone. Low ratings however appear to have a predictive value for relevant content. Discounting the negative information therefore does not only remove the irrelevant content from the top of the ranking, but also reduces the recall of relevant documents.