Increasingly, web recommender systems face scenarios where they need to serve suggestions to groups of users; for example, when families share e-commerce or movie rental web accounts. Research to date in this domain has proposed two approaches: computing recommendations for the group by merging any members' ratings into a single profile, or computing ranked recommendations for each individual that are then merged via a range of heuristics. In doing so, none of the past approaches reason on the preferences that arise in individuals when they are members of a group. In this work, we present a probabilistic framework, based on the notion of information matching, for group recommendation. This model defines group relevance as a combination of the item's relevance to each user as an individual and as a member of the group; it can then seamlessly incorporate any group recommendation strategy in order to rank items for a set of individuals. We evaluate the model's efficacy at generating recommendations for both single individuals and groups using the MovieLens and MoviePilot data sets. In both cases, we compare our results with baselines and state-of-the-art collaborative filtering algorithms, and show that the model outperforms all others over a variety of ranking metrics.

In this work, we propose a theory for information matching. It is motivated by the observation that retrieval is about the relevance matching between two sets of properties (features), namely, the information need representation and information item representation. However, many probabilistic retrieval models rely on fixing one representation and optimizing the other (e.g. fixing the single information need and tuning the document) but not both. Therefore, it is difficult to use the available related information on both the document and the query at the same time in calculating the probability of relevance. In this paper, we address the problem by hypothesizing the relevance as a logical relationship between the two sets of properties; the relationship is defined on two separate mappings between these properties. By using the hypothesis we develop a unified probabilistic relevance model which is capable of using all the available information. We validate the proposed theory by formulating and developing probabilistic relevance ranking functions for both ad-hoc text retrieval and collaborative filtering. Our derivation in text retrieval illustrates the use of the theory in the situation where no relevance information is available. In collaborative filtering, we show that the resulting recommender model unifies the user and item information into a relevance ranking function without applying any dimensionality reduction techniques or computing explicit similarity between two different users (or items), in contrast to the state-of-the-art recommender models.

In this paper, an Eliteness Hypothesis for information retrieval is proposed, where we define two generative processes to create information items and queries. By assuming the deterministic relationships between the eliteness of terms and relevance, we obtain a new theoretical retrieval framework. The resulting ranking function is a unified one as it is capable of using available relevance information on both the document and the query, which is otherwise unachievable by existing retrieval models. Our preliminary experiment on a simple ranking function has demonstrated the potential of the approach.

The structure of a sentence can be seen as a spanning tree in a linguistically augmented graph of syntactic nodes. This paper presents an approach for unlabeled dependency parsing based on this view. The first step involves marking the chunks and the chunk heads of a given sentence and then identifying the intra-chunk dependency relations. The second step involves learning to identify the inter-chunk dependency relations. For this, we use an initialization technique based on a measure we call Normalized Conditional Mutual Information (NCMI), in addition to a few linguistic constraints. We present the results for Hindi. We have achieved a precision of 80.83% for sentences of size less than 10 words and 66.71% overall. This is significantly better than the baseline in which random initialization is used.

Much work has been done on building a parser for natural languages, but most of this work has concentrated on supervised parsing. Unsupervised parsing is a less explored area, and unsupervised dependency parser has hardly been tried. In this paper we present two approaches for building an unsupervised dependency parser. One approach is based on learning dependency relations and the other on learning subtrees. We also propose some other applications of these approaches.

Text on the Web is available in numerous languages and encodings, often not according to any standards. The number of multilingual documents on the Web is also increasing. The problem of identifying the languages and encodings in a multilingual document and marking portions of a document with them has not been addressed so far. We present an exploration of this problem, the implied or required assumptions, and a solution. The problem can be divided into three parts: monolingual identification, enumeration of languages and identification of the language of every portion. For enumeration, we have been able to get a precision of 96.20%. We also experimented on language identification of each word. Given correct enumeration, we could obtain type precision of 90.91% and token precision of 86.80%. Finally, we show how precision is affected by language distance.

Text on the Web is available in numerous languages and encodings, often not according to any standards. The number of multilingual documents on the Web is also increasing. The problem of identifying the languages and encodings in a multilingual document and marking portions of a document with them has not been addressed so far. We present an exploration of this problem, the implied or required assumptions, and a solution. The problem can be divided into three parts: monolingual identification, enumeration of languages and identification of the language of every portion. For enumeration, we have been able to get a precision of 96.20%. We also experimented on language identification of each word. Given correct enumeration, we could obtain type precision of 90.91% and token precision of 86.80%. Finally, we show how precision is affected by language distance.

All languages mark tense, aspect and modality (TAM) in some way, but the markers don’t have a one-to-one mapping across languages. Many errors in machine translation (MT) are due to wrong translation of TAM markers. Reducing them can improve the performance of an MT sys-tem. We used about 9000 sentence pairs from an English-Hindi parallel corpus. These were manually annotated with TAM markers and their mappings. Based on this corpus, we identify the factors responsible for ambiguity in translation. We present the results for learning TAM marker translation using CRF. We achieved an improvement of 17.88% over the baseline.