Self-publication of news on Web sites is becoming a common application platform to enable more engaging interaction among users. Discussion in the form of comments following news postings can be effectively facilitated if the service provider can recommend articles based on not only the original news itself but also the thread of changing comments. This turns the traditional news recommendation to a "discussion moderator" that can intelligently assist online forums. In this work, we present a framework to implement such adaptive news recommendation. In addition, to alleviate the problem of recommending essentially identical articles, the relationship (duplication, generalization or specialization) between suggested news articles and the original posting is investigated. Experiments indicate that our proposed solutions provide an enhanced news recommendation service in forum-based social media.

With the development of Web 2.0, sentiment analysis has now become a popular research problem to tackle. Recently, topic models have been introduced for the simultaneous analysis for topics and the sentiment in a document. These studies, which jointly model topic and sentiment, take the advantage of the relationship between topics and sentiment, and are shown to be superior to traditional sentiment analysis tools. However, most of them make the assumption that, given the parameters, the sentiments of the words in the document are all independent. In our observation, in contrast, sentiments are expressed in a coherent way. The local conjunctive words, such as “and” or “but”, are often indicative of sentiment transitions. In this paper, we propose a major departure from the previous approaches by making two linked contributions. First, we assume that the sentiments are related to the topic in the document, and put forward a joint sentiment and topic model, i.e. Sentiment-LDA. Second, we observe that sentiments are dependent on local context. Thus, we further extend the Sentiment-LDA model to Dependency-Sentiment-LDA model by relaxing the sentiment independent assumption in Sentiment-LDA. The sentiments of words are viewed as a Markov chain in Dependency-Sentiment-LDA. Through experiments, we show that exploiting the sentiment dependency is clearly advantageous, and that the Dependency-Sentiment-LDA is an effective approach for sentiment analysis.

We consider the problem of automatic vocal melody transcription: translating an audio recording of a sung melody into a musical score. While previous work has focused on finding the closest notes to the singer's tracked pitch, we instead seek to recover the melody the singer intended to sing. Often, the melody a singer intended to sing differs from what they actually sang; our hypothesis is that this occurs in a singer-specific way. For example, a given singer may often be flat in certain parts of her range, or another may have difficulty with certain intervals. We thus pursue methods for singer-specific training which use learning to combine different methods for pitch prediction. In our experiments with human subjects, we show that via a short training procedure we can learn a singer-specific pitch predictor and significantly improve transcription of intended pitch over other methods. For an average user, our method gives a 20 to 30 percent reduction in pitch classification errors with respect to a baseline method which is comparable to commercial voice transcription tools. For some users, we achieve even more dramatic reductions. Our best results come from a combination of singer-specific-learning with non-singer-specific feature selection. We also discuss the implications of our work for training more general control signals. We make our experimental data available to allow others to replicate or extend our results.

Matrix factorization is a fundamental technique in machine learning that is applicable to collaborative filtering, information retrieval and many other areas. In collaborative filtering and many other tasks, the objective is to fill in missing elements of a sparse data matrix. One of the biggest challenges in this case is filling in a column or row of the matrix with very few observations. In this paper we introduce a Bayesian matrix factorization model that performs regression against side information known about the data in addition to the observations. The side information helps by adding observed entries to the factored matrices. We also introduce a nonparametric mixture model for the prior of the rows and columns of the factored matrices that gives a different regularization for each latent class. Besides providing a richer prior, the posterior distribution of mixture assignments reveals the latent classes. Using Gibbs sampling for inference, we apply our model to the Netflix Prize problem of predicting movie ratings given an incomplete user-movie ratings matrix. Incorporating rating information with gathered metadata information, our Bayesian approach outperforms other matrix factorization techniques even when using fewer dimensions.

Collaborative filtering systems have been developed to manage information overload and improve discourse in online communities. In such systems, users rank content provided by other users on the validity or usefulness within their particular context. The goal is that "good" content will rise to prominence and "bad" content will fade into obscurity. These filtering mechanisms are not well-understood and have known weaknesses. For example, they depend on the presence of a large crowd to rate content, but such a crowd may not be present. Additionally, the community's decisions determine which voices will reach a large audience and which will be silenced, but it is not known if these decisions represent "the wisdom of crowds" or a "censoring mob." Our approach uses statistical machine learning to predict community ratings. By extracting features that replicate the community's verdict, we can better understand collaborative filtering, improve the way the community uses the ratings of their members, and design agents that augment community decision-making. Slashdot is an example of such a community where peers will rate each others' comments based on their relevance to the post. This work extracts a wide variety of features from the Slashdot metadata and posts' linguistic contents to identify features that can predict the community rating. We find that author reputation, use of pronouns, and author sentiment are salient. We achieve 76% accuracy predicting community ratings as good, neutral, or bad.

We have been exploring the use of Web-derived knowledge bases through the development of Wikitology — a hybrid knowledge base of structured and unstructured information extracted from Wikipedia augmented by RDF data from DBpedia and other Linked Open Data resources. In this paper, we describe approaches that aid in enriching Wikipedia and thus the resources that derive from Wikipedia such as the Wikitology knowledge base, DBpedia, Freebase and Powerset.

Increasingly, we need to computationally understand real-time streams of information in places such as news feeds, speech streams, and social networks. We present Streaming AnalogySpace, an efficient technique that discovers correlations in and makes predictions about sparse natural-language data that arrives in a real-time stream. AnalogySpace is a noise-resistant PCA-based inference technique designed for use with collaboratively collected common sense knowledge and semantic networks. Streaming AnalogySpace advances this work by computing it incrementally using CCIPCA, and keeping a dense cache of recently-used features to efficiently represent a sparse and open domain. We show that Streaming AnalogySpace converges to the results of standard AnalogySpace, and verify this by evaluating its accuracy empirically on common-sense predictions against standard AnalogySpace.

Music composition used to be a pen and paper activity. These these days music is often composed with the aid of computer software, even to the point where the computer compose parts of the score autonomously. The composition of most styles of music is governed by rules. We show that by approaching the automation, analysis and verification of composition as a knowledge representation task and formalising these rules in a suitable logical language, powerful and expressive intelligent composition tools can be easily built. This application paper describes the use of answer set programming to construct an automated system, named ANTON, that can compose melodic, harmonic and rhythmic music, diagnose errors in human compositions and serve as a computer-aided composition tool. The combination of harmonic, rhythmic and melodic composition in a single framework makes ANTON unique in the growing area of algorithmic composition. With near real-time composition, ANTON reaches the point where it can not only be used as a component in an interactive composition tool but also has the potential for live performances and concerts or automatically generated background music in a variety of applications. With the use of a fully declarative language and an "off-the-shelf" reasoning engine, ANTON provides the human composer a tool which is significantly simpler, more compact and more versatile than other existing systems. This paper has been accepted for publication in Theory and Practice of Logic Programming (TPLP).

Cross-document coreference, the problem of resolving entity mentions across multi-document collections, is crucial to automated knowledge base construction and data mining tasks. However, the scarcity of large labeled data sets has hindered supervised machine learning research for this task. In this paper we develop and demonstrate an approach based on ``distantly-labeling'' a data set from which we can train a discriminative cross-document coreference model. In particular we build a dataset of more than a million people mentions extracted from 3.5 years of New York Times articles, leverage Wikipedia for distant labeling with a generative model (and measure the reliability of such labeling); then we train and evaluate a conditional random field coreference model that has factors on cross-document entities as well as mention-pairs. This coreference model obtains high accuracy in resolving mentions and entities that are not present in the training data, indicating applicability to non-Wikipedia data. Given the large amount of data, our work is also an exercise demonstrating the scalability of our approach.

In this paper we examine the effects of noise when creating a real-world weblog corpus for information retrieval. We focus on the DiffPost (Lee et al. 2008) approach to noise removal from blog pages, examining the difficulties encountered when crawling the blogosphere during the creation of a real-world corpus of blog pages. We introduce and evaluate a number of enhancements to the original DiffPost approach in order to increase the robustness of the algorithm. We then extend DiffPost by looking at the anchor-text to text ratio, and discover that the time-interval between crawls is more important to the successful application of noise-removal algorithms within the blog context, than any additional improvements to the removal algorithm itself.