Shengbo Guo

Probabilistic latent maximal marginal relevance

Diversity has been heavily motivated in the information retrieval literature as an objective criterion for result sets in search and recommender systems. Perhaps one of the most well-known and most used algorithms for result set diversification is that of Maximal Marginal Relevance (MMR). In this paper, we show that while MMR is somewhat ad-hoc and motivated from a purely pragmatic perspective, we can derive a more principled variant via probabilistic inference in a latent variable graphical model. This novel derivation presents a formal probabilistic latent view of MMR (PLMMR) that (a) removes the need to manually balance relevance and diversity parameters, (b) shows that specific definitions of relevance and diversity metrics appropriate to MMR emerge naturally, and (c) formally derives variants of latent semantic indexing (LSI) similarity metrics for use in PLMMR. Empirically, PLMMR outperforms MMR with standard term frequency based similarity and diversity metrics since PLMMR maximizes latent diversity in the results.

Connecting comments and tags: improved modeling of social tagging systems

Collaborative tagging systems are now deployed extensively to help users share and organize resources. Tag prediction and recommendation can simplify and streamline the user experience, and by modeling user preferences, predictive accuracy can be significantly improved. However, previous methods typically model user behavior based only on a log of prior tags, neglecting other behaviors and information in social tagging systems, e.g., commenting on items and connecting with other users. On the other hand, little is known about the connection and correlations among these behaviors and contexts in social tagging systems. In this paper, we investigate improved modeling for predictive social tagging systems. Our explanatory analyses demonstrate three significant challenges: coupled high order interaction, data sparsity and cold start on items. We tackle these problems by using a generalized latent factor model and fully Bayesian treatment. To evaluate performance, we test on two real-world data sets from Flickr and Bibsonomy. Our experiments on these data sets show that to achieve best predictive performance, it is necessary to employ a fully Bayesian treatment in modeling high order relations in social tagging system. Our methods noticeably outperform state-of-the-art approaches.

Diverse retrieval via greedy optimization of expected 1-call@k in a latent subtopic relevance model

It has been previously observed that optimization of the 1-call@k relevance objective (i.e., a set-based objective that is 1 if at least one document is relevant, otherwise 0) empirically correlates with diverse retrieval. In this paper, we proceed one step further and show theoretically that greedily optimizing expected 1-call@k w.r.t. a latent subtopic model of binary relevance leads to a diverse retrieval algorithm sharing many features of existing diversification approaches. This new result is complementary to a variety of diverse retrieval algorithms derived from alternate rank-based relevance criteria such as average precision and reciprocal rank. As such, the derivation presented here for expected 1-call@k provides a novel theoretical perspective on the emergence of diversity via a latent subtopic model of relevance --- an idea underlying both ambiguous and faceted subtopic retrieval that have been used to motivate diverse retrieval.

Gene selection based on mutual information for the classification of multi-class cancer

With the development of mirocarray technology, microarray data are widely used in the diagnoses of cancer subtypes. However, people are still facing the complicated problem of accurate diagnosis of cancer subtypes. Building classifiers based on the selected key genes from microarray data is a promising approach for the development of microarray technology; yet the selection of non-redundant but relevant genes is complicated. The selected genes should be small enough to allow diagnosis even in regular laboratories and ideally identify genes involved in cancer-specific regulatory pathways. Instead of the traditional gene selection methods used for the classification of two categories of cancers, in the present paper, a novel gene selection algorithm based on mutual information is proposed for the classification of multi-class cancer using microarray data, and the selected key genes are fed into the classifier to classify the cancer subtypes. In our algorithm, mutual information is employed to select key genes related with class distinction. The application on the breast cancer data suggests that the present algorithm can identify the key genes to the BRCA1 mutations/BRCA2 mutations/the sporadic mutations class distinction since the result of our proposed algorithm is promising, because our method can perform the classification of the three types of breast cancer effectively and efficiently. And two more microarray datasets, leukemia and ovarian cancer data, are also employed to validate the performance of our method. The performances of these applications demonstrate the high quality of our method. Based on the present work, our method can be widely used to discriminate different cancer subtypes, which will contribute to the development of technology for the recovery of the cancer.

Score-Based bayesian skill learning

We extend the Bayesian skill rating system of TrueSkill to accommodate score-based match outcomes. TrueSkill has proven to be a very effective algorithm for matchmaking -- the process of pairing competitors based on similar skill-level -- in competitive online gaming. However, for the case of two teams/players, TrueSkill only learns from win, lose, or draw outcomes and cannot use additional match outcome information such as scores. To address this deficiency, we propose novel Bayesian graphical models as extensions of TrueSkill that (1) model players offence and defence skills separately and (2) model how these offence and defence skills interact to generate score-based match outcomes. We derive efficient (approximate) Bayesian inference methods for inferring latent skills in these new models and evaluate them on three real data sets including Halo 2 XBox Live matches. Empirical evaluations demonstrate that the new score-based models (a) provide more accurate win/loss probability estimates than TrueSkill when training data is limited, (b) provide competitive and often better win/loss classification performance than TrueSkill, and (c) provide reasonable score outcome predictions with an appropriate choice of likelihood -- prediction for which TrueSkill was not designed, but which can be useful in many applications.

On the mathematical relationship between expected n-call@k and the relevance vs. diversity trade-off

It has been previously noted that optimization of the <i>n</i>-call@<i>k</i> relevance objective (i.e., a set-based objective that is 1 if at least <i>n</i> documents in a set of <i>k</i> are relevant, otherwise 0) encourages more result set diversification for smaller <i>n</i>, but this statement has never been formally quantified. In this work, we explicitly derive the mathematical relationship between <i>expected n-call@k</i> and the <i>relevance vs. diversity trade-off</i> --- through fortuitous cancellations in the resulting combinatorial optimization, we show the trade-off is a simple and intuitive function of <i>n</i> (notably independent of the result set size <i>k</i> e <i>n</i>), where diversification increases as <i>n</i> approaches 1.

Multiattribute bayesian preference elicitation with pairwise comparison queries

Preference elicitation (PE) is an very important component of interactive decision support systems that aim to make optimal recommendations to users by actively querying their preferences In this paper, we present three principles important for PE in real-world problems: (1) multiattribute, (2) low cognitive load, and (3) robust to noise In light of three requirements, we introduce an approximate PE framework based on a variant of TrueSkill for performing efficient closed-form Bayesian updates and query selection for a multiattribute utility belief state — a novel PE approach that naturally facilitates the efficient evaluation of value of information (VOI) for use in query selection strategies Our VOI query strategy satisfies all three principles and performs on par with the most accurate algorithms on experiments with a synthetic data set.

Bark classification based on gabor filter features using RBPNN neural network

This paper proposed a new method of extracting texture features based on Gabor wavelet. In addition, the application of these features for bark classification applying radial basis probabilistic network (RBPNN) has been introduced. In this method, the bark texture feature is firstly extracted by filtering the image with different orientations and scales filters, then the mean and standard deviation of the image output are computed, the image which have been filtered in the frequency domain. Finally, the obtained Gabor feature vectors are fed up into RBPNN for classification. Experimental results show that, first, features extracted using the proposed approach can be used for bark texture classification. Second, compared with radial basis function neural network (RBFNN), the RBPNN achieves higher recognition rate and better classification efficiency when the feature vectors have low-dimensions.

Long-Range Temporal Correlations in the Spontaneous in vivo Activity of Interneuron in the Mouse Hippocampus

The spontaneous in vivo firings of neuron in mouse hippocampus are generally considered as neuronal noise, where there is no any correlation in the inter-spike interval (ISI) sequences. In the present study, we investigate the nature of the ISI sequences of neuron in CA1 area of mouse hippocampus. By using the detrended fluctuation analysis (DFA), we calculated the fluctuation or scaling exponent of the ISI sequences. The results indicated that there exists the long-range power-law correlation over large time scale in the ISI sequences. To further investigate the long-range correlation of ISI, we studied the long-range correlation of ISI sequences from different types of neurons in mouse hippocampus, which are four types of interneurons categorized by their firing patterns. Our results show the presence of long-range correlations in the ISI sequence of different types of neurons. Furthermore, the shuffle surrogate data achieved by randomly shuffle the original ISI sequence is used to verify our conclusion. The application of shuffle surrogate shows that the long-range correlation is destroyed by randomly shuffle, which demonstrates that there is actually the long-range correlation in the ISI sequence. Furthermore, we also compare the long-range correlations of ISI sequence when mice are in different behavioral states, slow-wave sleep (SWS) and active exploration (AE). Our results indicated that the ISI sequences exhibit different extent of long-range correlations: the long-range correlation is significantly stronger when mice are in AE than that of ISI sequence when mice are in SWS, which demonstrated that the varied long-range correlations exhibiting in ISIs of interneurons might be associated with activities of neuronal network regulating the ongoing neuronal activity of different interneurons.

Choosing Business Collaborators Using Computing Intelligence Methods

Inter-firm collaboration has become a common feature of the developing international economy. Firms as well as the nations have more relationships with each other. Even relatively closed economies or industries are becoming more open, Australia and China are examples of this case. The benefits generated from collaboration and the motivations to form collaboration are investigated by some researchers. However, the widely studied relationships between collaboration and profits are based on tangible assets and turnovers whereas most intangible assets and benefits are neglected during the economic analysis. In the present paper, two methods, naive Bayes and neural network, from computing intelligence are used to study the benefits acquired from collaboration. These two methods are used to learn the relationship and make prediction for a specified collaboration. The proposed method has been applied to a practical case of WEMOSOFT, an independent development department under MOBOT. The predication accuracies are 87.18% and 92.31%, for neural network and naive Bayes, respectively. Experimental result demonstrates that the proposed method is an effective and efficient way to prediction the benefit of collaboration and choose the appropriate collaborator.