David Vallet

Personalizing web search with folksonomy-based user and document profiles

Web search personalization aims to adapt search results to a user based on his tastes, interests and needs. The way in which such personal preferences are captured, modeled and exploited distinguishes the different personalization strategies. In this paper, we propose to represent a user profile in terms of social tags, manually provided by users in folksonomy systems to describe, categorize and organize items of interest, and investigate a number of novel techniques that exploit the users’ social tags to re-rank results obtained with a Web search engine. An evaluation conducted with a dataset from Delicious social bookmarking system shows that our personalization techniques clearly outperform state of the art approaches.

An asynchronous collaborative search system for online video search

There are a number of multimedia tasks and environments that can be collaborative in nature and involve contributions from more than one individual. Examples of such tasks include organising photographs or videos from multiple people from a large event, students working together to complete a class project, or artists and/or animators working on a production. Despite this, current state of the art applications that have been created to assist in multimedia search and organisation focus on a single user searching alone and do not take into consideration the collaborative nature of a large number of multimedia tasks. The limited work in collaborative search for multimedia applications has concentrated mostly on synchronous, and quite often co-located, collaboration between persons. However, these collaborative scenarios are not always practical or feasible. In order to overcome these shortcomings we have created an innovative system for online video search, which provides mechanisms for groups of users to collaborate both asynchronously and remotely on video search tasks. In order to evaluate our system an user evaluation was conducted. This evaluation simulated multiple conditions and scenarios for collaboration, varying on awareness, division of labour, sense making and persistence. The outcome of this evaluation demonstrates the benefit and usability of our system for asynchronous and remote collaboration between users. In addition the results of this evaluation provide a comparison between implicit and explicit collaboration in the same search system.

Characterizing and Predicting Viral-and-Popular Video Content

The proliferation of online video content has triggered numerous works on its evolution and popularity, as well as on the effect of social sharing on content propagation. In this paper, we focus on the observable dependencies between the virality of video content on a micro-blogging social network (in this case, Twitter) and the popularity of such content on a video distribution service (YouTube). To this end, we collected and analysed a corpus of Twitter posts containing links to YouTube clips and the corresponding video meta-data from YouTube. Our analysis highlights the unique properties of content that is both popular and viral, which allows such content to attract high number of views on YouTube and achieve fast propagation on Twitter. With this in mind, we proceed to the predictions of popular-and-viral clips and propose a framework that can, with high degree of accuracy and low amount of training data, predict videos that are likely to be popular, viral, and both. The key contribution of our work is the focus on cross-system dynamics between YouTube and Twitter. We conjecture and validate that cross-system prediction of both popularity and virality of videos is feasible, and can be performed with a reasonably high degree of accuracy. One of our key findings is that YouTube features capturing user engagement, have strong virality prediction capabilities. This findings allows to solely rely on data extracted from a video sharing service to predict popularity and virality aspects of videos.

Improving business rating predictions using graph based features

Many types of recommender systems rely on a rich ensemble of user, item, and context features when generating recommendations for users. The features can be either manually engineered or automatically extracted from the available data, such that feature engineering becomes an important step in the recommendation process. In this work, we propose to leverage graph based representation of the data in order to generate and automatically populate features. We represent the standard user-item rating matrix and some domain metadata, as graph vertices and edges. Then, we apply a suite of graph theory and network analysis metrics to the graph based data representation, to populate features that augment the original user-item ratings data. The augmented data is fed into a classifier that predicts unknown user ratings, which are used for the generation of recommendations. We evaluate the proposed methodology using the recently released Yelp business ratings dataset. Our results indicate that the automatically populated graph features allow for more accurate and robust predictions, with respect to both the variability and sparsity of ratings.

ViGOR: a grouping oriented interface for search and retrieval in video libraries

In this paper, we present ViGOR (Video Grouping, Organisation and Retrieval) a video retrieval system that allows users to group videos in order to facilitate video retrieval tasks. In this way users are able to visualise and conceptualise many aspects of their search tasks and carry out a localised search in order to solve a more global search problem. The main objective of this work is to aid users while carrying out explorative video retrieval tasks; these tasks can be often ambiguous and multi-faceted. Two user evaluations were carried out in order to evaluate the usefulness of this grouping paradigm for assisting users. The first evaluation involved users carrying out broad tasks on YouTube, and gave insights into the application of our interface to a vast online video collection. The second evaluation involved users carrying out focused tasks on the TRECVID 2007 video collection, allowing a comparison over a local collection, on which we could extract a number of content-based features. The results of our evaluations show that the use of the ViGOR system results in an increase in user performance and user satisfaction, showing the potential of a grouping paradigm for video search for various tasks in a variety of diverse video collections.

Examining Additivity and Weak Baselines

We present a study of which baseline to use when testing a new retrieval technique. In contrast to past work, we show that measuring a statistically significant improvement over a weak baseline is not a good predictor of whether a similar improvement will be measured on a strong baseline. Sometimes strong baselines are made worse when a new technique is applied. We investigate whether conducting comparisons against a range of weaker baselines can increase confidence that an observed effect will also show improvements on a stronger baseline. Our results indicate that this is not the case -- at best, testing against a range of baselines means that an experimenter can be more confident that the new technique is unlikely to significantly harm a strong baseline. Examining recent past work, we present evidence that the information retrieval (IR) community continues to test against weak baselines. This is unfortunate as, in light of our experiments, we conclude that the only way to be confident that a new technique is a contribution is to compare it against nothing less than the state of the art.

Using score differences for search result diversification

We investigate the application of a light-weight approach to result list clustering for the purposes of diversifying search results. We introduce a novel post-retrieval approach, which is independent of external information or even the full-text content of retrieved documents; only the retrieval score of a document is used. Our experiments show that this novel approach is beneficial to effectiveness, albeit only on certain baseline systems. The fact that the method works indicates that the retrieval score is potentially exploitable in diversity.

Predicting Re-finding Activity and Difficulty

In this study, we address the problem of identifying if users are attempting to re-find information and estimating the level of difficulty of the re-finding task. We propose to consider the task information (e.g. multiple queries and click information) rather than only queries. Our resultant prediction models are shown to be significantly more accurate (by 2%) than the current state of the art. While past research assumes that previous search history of the user is available to the prediction model, we examine if re-finding detection is possible without access to this information. Our evaluation indicates that such detection is possible, but more challenging. We further describe the first predictive model in detecting re-finding difficulty, showing it to be significantly better than existing approaches for detecting general search difficulty.

Graph-Based Recommendations: Make the Most Out of Social Data

Recommender systems use nowadays more and more data about users and items as part of the recommendation process. The availability of auxiliary data, going beyond the mere user/item data, has the potential to improve recommendations. In this work we examine the contribution of two types of social auxiliary data – namely, tags and friendship links – to the accuracy of a graph-based recommender. We measure the impact of the availability of auxiliary data on the recommendations using features extracted from both the auxiliary and the original data. The evaluation shows that the social auxiliary data improves the accuracy of the recommendations, and that the greatest improvement is achieved when graph features mirroring the nature of the auxiliary data are extracted by the recommender.

Exploiting Social Tagging Profiles to Personalize Web Search

In this paper, we investigate the exploitation of user profiles defined in social tagging services to personalize Web search. One of the key challenges of a personalization framework is the elicitation of user profiles able to represent user interests. We propose a personalization approach that exploits the tagging information of users within a social tagging service as a way of obtaining their interests. We evaluate this approach in Delicious, a social Web bookmarking service, and apply our personalization approach to a Web search system. Our evaluation results indicate a clear improvement of our approach over related state of the art personalization approaches.