LRD : Latent Relation Discovery for vector space expansion and information retrieval

In this paper, we propose a text mining method called LRD (latent relation discovery), which extends the traditional vector space model of document representation in order to improve information retrieval (IR) on documents and document clustering. Our LRD method extracts terms and entities, such as person, organization, or project names, and discovers relationships between them by taking into account their co-occurrence in textual corpora. Given a target entity, LRD discovers other entities closely related to the target effectively and efficiently. With respect to such relatedness, a measure of relation strength between entities is defined. LRD uses relation strength to enhance the vector space model, and uses the enhanced vector space model for query based IR on documents and clustering documents in order to discover complex relationships among terms and entities. Our experiments on a standard dataset for query based IR shows that our LRD method performed significantly better than traditional vector space model and other five standard statistical methods for vector expansion.

Biomedical events extraction using the hidden vector state model

Objective: Biomedical events extraction concerns about events describing changes on the state of bio-molecules from literature. Comparing to the protein-protein interactions (PPIs) extraction task which often only involves the extraction of binary relations between two proteins, biomedical events extraction is much harder since it needs to deal with complex events consisting of embedded or hierarchical relations among proteins, events, and their textual triggers. In this paper, we propose an information extraction system based on the hidden vector state (HVS) model, called HVS-BioEvent, for biomedical events extraction, and investigate its capability in extracting complex events. Methods and material: HVS has been previously employed for extracting PPIs. In HVS-BioEvent, we propose an automated way to generate abstract annotations for HVS training and further propose novel machine learning approaches for event trigger words identification, and for biomedical events extraction from the HVS parse results. Results: Our proposed system achieves an F-score of 49.57% on the corpus used in the BioNLP'09 shared task, which is only 2.38% lower than the best performing system by UTurku in the BioNLP'09 shared task. Nevertheless, HVS-BioEvent outperforms UTurku's system on complex events extraction with 36.57% vs. 30.52% being achieved for extracting regulation events, and 40.61% vs. 38.99% for negative regulation events. Conclusions: The results suggest that the HVS model with the hierarchical hidden state structure is indeed more suitable for complex event extraction since it could naturally model embedded structural context in sentences.

Extracting protein-protein interactions from MEDLINE using the hidden vector state model

A major challenge in text mining for biomedicine is automatically extracting protein-protein interactions from the vast amount of biomedical literature. We have constructed an information extraction system based on the Hidden Vector State (HVS) model for protein-protein interactions. The HVS model can be trained using only lightly annotated data whilst simultaneously retaining sufficient ability to capture the hierarchical structure. When applied in extracting protein-protein interactions, we found that it performed better than other established statistical methods and achieved 61.5% in F-score with balanced recall and precision values. Moreover, the statistical nature of the pure data-driven HVS model makes it intrinsically robust and it can be easily adapted to other domains.

Discriminative training of the hidden vector state model for semantic parsing

In this paper, we discuss how discriminative training can be applied to the hidden vector state (HVS) model in different task domains. The HVS model is a discrete hidden Markov model (HMM) in which each HMM state represents the state of a push-down automaton with a finite stack size. In previous applications, maximum-likelihood estimation (MLE) is used to derive the parameters of the HVS model. However, MLE makes a number of assumptions and unfortunately some of these assumptions do not hold. Discriminative training, without making such assumptions, can improve the performance of the HVS model by discriminating the correct hypothesis from the competing hypotheses. Experiments have been conducted in two domains: the travel domain for the semantic parsing task using the DARPA Communicator data and the Air Travel Information Services (ATIS) data and the bioinformatics domain for the information extraction task using the GENIA corpus. The results demonstrate modest improvements of the performance of the HVS model using discriminative training. In the travel domain, discriminative training of the HVS model gives a relative error reduction rate of 31 percent in F-measure when compared with MLE on the DARPA Communicator data and 9 percent on the ATIS data. In the bioinformatics domain, a relative error reduction rate of 4 percent in F-measure is achieved on the GENIA corpus.

Ontology-based protein-protein interactions extraction from literature using the hidden vector state model

This paper proposes a novel framework of incorporating protein-protein interactions (PPI) ontology knowledge into PPI extraction from biomedical literature in order to address the emerging challenges of deep natural language understanding. It is built upon the existing work on relation extraction using the Hidden Vector State (HVS) model. The HVS model belongs to the category of statistical learning methods. It can be trained directly from un-annotated data in a constrained way whilst at the same time being able to capture the underlying named entity relationships. However, it is difficult to incorporate background knowledge or non-local information into the HVS model. This paper proposes to represent the HVS model as a conditionally trained undirected graphical model in which non-local features derived from PPI ontology through inference would be easily incorporated. The seamless fusion of ontology inference with statistical learning produces a new paradigm to information extraction.

Probabilistic resource space model for managing resources in cyber-physical society

Classification is the most basic method for organizing resources in the physical space, cyber space, socio space and mental space. To create a unified model that can effectively manage resources in different spaces is a challenge. The Resource Space Model RSM is to manage versatile resources with a multi-dimensional classification space. It supports generalization and specialization on multi-dimensional classifications. This paper introduces the basic concepts of RSM, and proposes the Probabilistic Resource Space Model, P-RSM, to deal with uncertainty in managing various resources in different spaces of the cyber-physical society. P-RSM’s normal forms, operations and integrity constraints are developed to support effective management of the resource space. Characteristics of the P-RSM are analyzed through experiments. This model also enables various services to be described, discovered and composed from multiple dimensions and abstraction levels with normal form and integrity guarantees. Some extensions and applications of the P-RSM are introduced.

Clustering web documents using hierarchical representation with multi-granularity

Web document cluster analysis plays an important role in information retrieval by organizing large amounts of documents into a small number of meaningful clusters. Traditional web document clustering is based on the Vector Space Model (VSM), which takes into account only two-level (document and term) knowledge granularity but ignores the bridging paragraph granularity. However, this two-level granularity may lead to unsatisfactory clustering results with “false correlation”. In order to deal with the problem, a Hierarchical Representation Model with Multi-granularity (HRMM), which consists of five-layer representation of data and a twophase clustering process is proposed based on granular computing and article structure theory. To deal with the zero-valued similarity problemresulted from the sparse term-paragraphmatrix, an ontology based strategy and a tolerance-rough-set based strategy are introduced into HRMM. By using granular computing, structural knowledge hidden in documents can be more efficiently and effectively captured in HRMM and thus web document clusters with higher quality can be generated. Extensive experiments show that HRMM, HRMM with tolerancerough-set strategy, and HRMM with ontology all outperform VSM and a representative non VSM-based algorithm, WFP, significantly in terms of the F-Score.

Modelling financial time series with switching state space models

The deficiencies of stationary models applied to financial time series are well documented. A special form of non-stationarity, where the underlying generator switches between (approximately) stationary regimes, seems particularly appropriate for financial markets. We use a dynamic switching (modelled by a hidden Markov model) combined with a linear dynamical system in a hybrid switching state space model (SSSM) and discuss the practical details of training such models with a variational EM algorithm due to [Ghahramani and Hilton,1998]. The performance of the SSSM is evaluated on several financial data sets and it is shown to improve on a number of existing benchmark methods.

A novel framework of training hidden Markov support vector machines from lightly-annotated data

Natural language understanding (NLU) aims to map sentences to their semantic mean representations. Statistical approaches to NLU normally require fully-annotated training data where each sentence is paired with its word-level semantic annotations. In this paper, we propose a novel learning framework which trains the Hidden Markov Support Vector Machines (HM-SVMs) without the use of expensive fully-annotated data. In particular, our learning approach takes as input a training set of sentences labeled with abstract semantic annotations encoding underlying embedded structural relations and automatically induces derivation rules that map sentences to their semantic meaning representations. The proposed approach has been tested on the DARPA Communicator Data and achieved 93.18% in F-measure, which outperforms the previously proposed approaches of training the hidden vector state model or conditional random fields from unaligned data, with a relative error reduction rate of 43.3% and 10.6% being achieved.

A framework for automated construction of resource space based on background knowledge

Resource Space Model is a kind of data model which can effectively and flexibly manage the digital resources in cyber-physical system from multidimensional and hierarchical perspectives. This paper focuses on constructing resource space automatically. We propose a framework that organizes a set of digital resources according to different semantic dimensions combining human background knowledge in WordNet and Wikipedia. The construction process includes four steps: extracting candidate keywords, building semantic graphs, detecting semantic communities and generating resource space. An unsupervised statistical language topic model (i.e., Latent Dirichlet Allocation) is applied to extract candidate keywords of the facets. To better interpret meanings of the facets found by LDA, we map the keywords to Wikipedia concepts, calculate word relatedness using WordNet's noun synsets and construct corresponding semantic graphs. Moreover, semantic communities are identified by GN algorithm. After extracting candidate axes based on Wikipedia concept hierarchy, the final axes of resource space are sorted and picked out through three different ranking strategies. The experimental results demonstrate that the proposed framework can organize resources automatically and effectively.©2013 Published by Elsevier Ltd. All rights reserved.

Low level feature detection in human vision

Influential models of edge detection have generally supposed that an edge is detected at peaks in the 1st derivative of the luminance profile, or at zero-crossings in the 2nd derivative. However, when presented with blurred triangle-wave images, observers consistently marked edges not at these locations, but at peaks in the 3rd derivative. This new phenomenon, termed ‘Mach edges’ persisted when a luminance ramp was added to the blurred triangle-wave. Modelling of these Mach edge detection data required the addition of a physiologically plausible filter, prior to the 3rd derivative computation. A viable alternative model was examined, on the basis of data obtained with short-duration, high spatial-frequency stimuli. Detection and feature-making methods were used to examine the perception of Mach bands in an image set that spanned a range of Mach band detectabilities. A scale-space model that computed edge and bar features in parallel provided a better fit to the data than 4 competing models that combined information across scale in a different manner, or computed edge or bar features at a single scale. The perception of luminance bars was examined in 2 experiments. Data for one image-set suggested a simple rule for perception of a small Gaussian bar on a larger inverted Gaussian bar background. In previous research, discriminability (d’) has typically been reported to be a power function of contrast, where the exponent (p) is 2 to 3. However, using bar, grating, and Gaussian edge stimuli, with several methodologies, values of p were obtained that ranged from 1 to 1.7 across 6 experiments. This novel finding was explained by appealing to low stimulus uncertainty, or a near-linear transducer.

Protein-protein interactions classification from text via local learning with class priors

Text classification is essential for narrowing down the number of documents relevant to a particular topic for further pursual, especially when searching through large biomedical databases. Protein-protein interactions are an example of such a topic with databases being devoted specifically to them. This paper proposed a semi-supervised learning algorithm via local learning with class priors (LL-CP) for biomedical text classification where unlabeled data points are classified in a vector space based on their proximity to labeled nodes. The algorithm has been evaluated on a corpus of biomedical documents to identify abstracts containing information about protein-protein interactions with promising results. Experimental results show that LL-CP outperforms the traditional semisupervised learning algorithms such as SVMand it also performs better than local learning without incorporating class priors.

Does online information drive offline revenues? Only for specific products and consumer segments!

While many offline retailers have developed informational websites that offer information on products and prices, the key question for such informational websites is whether they can increase revenues via web-to-store shopping. The current paper draws on the information search literature to specify and test hypotheses regarding the offline revenue impact of adding an informational website. Explicitly considering marketing efforts, a latent class model distinguishes consumer segments with different short-term revenue effects, while a Vector Autoregressive model on these segments reveals different long-term marketing response. We find that the offline revenue impact of the informational website critically depends on the product category and customer segment. The lower online search costs are especially beneficial for sensory products and for customers distant from the store. Moreover, offline revenues increase most for customers with high web visit frequency. We find that customers in some segments buy more and more expensive products, suggesting that online search and offline purchases are complements. In contrast, customers in a particular segment reduce their shopping trips, suggesting their online activities partially substitute for experiential shopping in the physical store. Hence, offline retailers should use specific online activities to target specific product categories and customer segments.

Bid-ask spread dynamics in foreign exchange markets

The aim of this paper is to examine the short term dynamics of foreign exchange rate spreads. Using a vector autoregressive model (VAR) we show that most of the variation in the spread comes from the long run dependencies between past and future spreads rather than being caused by changes in inventory, adverse selection, cost of carry or order processing costs. We apply the Integrated Cumulative Sum of Squares (ICSS) algorithm of Inclan and Tiao (1994) to discover how often spread volatility changes. We find that spread volatility shifts are relatively uncommon and shifts in one currency spread tend not to spillover to other currency spreads. © 2013.

The efficiency of multivariate macroeconomic forecasts

We examine the efficiency of multivariate macroeconomic forecasts by estimating a vector autoregressive model on the forecast revisions of four variables (GDP, inflation, unemployment and wages). Using a data set of professional forecasts for the G7 countries, we find evidence of cross‐series revision dynamics. Specifically, forecasts revisions are conditionally correlated to the lagged forecast revisions of other macroeconomic variables, and the sign of the correlation is as predicted by conventional economic theory. This indicates that forecasters are slow to incorporate news across variables. We show that this finding can be explained by forecast underreaction.