Dequan Zheng

Research on search results optimization technology with category features integration

The optimization of search results has always been the research hot spot in the area of search engine. In previous work, various kinds of document ranking were used to optimize the search results, in which topic partition by clustering has been proved to be a good way. However, the clusters, containing a lot of documents unorganized, still directly limit the retrieval speed. To address this issue, the paper firstly integrates the two methods together to re-rank the documents in clusters. We find that the category features, which have great discernibility for categories, have good effects on the document sequencing. Thereupon we attempt to apply the category features into search results on the basis of the clusters. Related experiments show that our Top N results are more in line with the users’ needs and the retrieval speed can be implicitly improved, which proves that our approach significantly outperforms the original clustering method.

Chinese Term Extraction Using Different Types of Relevance

This paper presents a new term extraction approach using relevance between term candidates calculated by a link analysis based method. Different types of relevance are used separately or jointly for term verification. The proposed approach requires no prior domain knowledge and no adaptation for new domains. Consequently, the method can be used in any domain corpus and it is especially useful for resource-limited domains. Evaluations conducted on two different domains for Chinese term extraction show significant improvements over existing techniques and also verify the efficiency and relative domain independent nature of the approach.

Key-Phrase Extraction Based on a Combination of CRF Model with Document Structure

Key-Phrase should not only reflect the main content of a document, but also reflect the specialty of this document. Key-Phrase extraction is an important technique in the field of text information processing. With the advent of the Internet age, on-line file shows an astonishing increase in geometry and information explosion has became the main character of this age. Searching and making use of network information becomes more difficult. Therefore, automatically extraction on keyword is required. This paper uses the idea of classification to complete the task of Key-Phrase extraction, which uses SVM to build classification model and uses CRF to extract Key-Phrases. The testing result shows that, the mentioned extraction approach has improved dramatically compared with previous methods in precision and recall rate.

Research on image classification based on a combination of text and visual features

As more and more text-image co-occurrence data become available on the web, mining on those data is playing an increasingly important role in web applications. In this paper, we consider utilizing description information to help image classification and propose a novel image classification method focusing on text-image co-occurrence data. In general, there are three main steps in our system: feature extraction, training classifiers and classifier fusion. In feature extraction phase, several features are extracted including not only visual features such as color, shape, texture, but also text features. In the process of training classifiers, visual and text classifiers are trained separately with SVM model. Finally, Weight learning is used to build the classifier fusion system. Comparing with other methods, we make full use of unstructured texts around images and filter text features through information gain, also efficient combination of features is achieved by comparing different combination methods. Experimental results show that our method is efficient and enhances the accuracy of image classification.

Product features mining based on Conditional Random Fields model

Opinion mining has become a hot issue attracting the attention of many researchers recently, in which the opinion feature is essential to its modeling. In the opinion mining of products, opinion feature identification is to mine product features from product reviews. In this paper, we present a Conditional Random Fields model based Chinese product features identification approach, integrating the chunk features and heuristic position information in addition to the word features, part-of-speech features and context features. Experiments show that the proposed techniques effectively improve the performance of product opinion mining.

Automatic Domain-Ontology Relation Extraction from Semi-structured Texts

This paper presents a new method to acquire Domain-Ontology relations from semi-structured data sources. First, obtain Web documents according to the co-occurrence of concept instance and attribute value. Further, define formats of relation patterns, and extract pattern instances from Web documents, including pattern clustering and pattern combining in each cluster. Finally, relation pattern instances are applied to gain attribute values of new concept instances in Domain-Ontology. Experiments are carried out in the field of film, the rate of pattern incorrect-division and pattern leakage are respectively 0.19% and 1.31%, the highest precision of combined relation patterns reaches 85%. Experimental results demonstrate that the method developed in this paper is fairly efficient.

Chinese-English Cross-Lingual Information Retrieval based on Domain Ontology Knowledge

For improving the effectiveness of cross-lingual information retrieval (CLIR), a domain ontology knowledge based method is presented to apply to C-E CLIR. In this study, the domain ontology knowledge is acquired from both source language user queries and target documents to select target translation and re-rank initial retrieval documents set. The C-E CLIR dataset from NTCIR-4 Workshop is used to evaluate the effectiveness of this method. Different from previous works, we make use of source language user queries in total C-E CLIR and compared with previous works, this method improved the precision

Chinese Microblog Sentiment Analysis Based on Semi-supervised Learning

This paper adopts a semi-supervised method which is based on bootstrapping to analyze Sina microblog data which size is about 269 M. The Support Vector Machine (SVM) method is used in subjective and objective classification and polarity classification. Our method can extend the size of seed samples by learning automatically with a small size of labeled corpus. It can improve the ability of sentiment classification of SVM by using the iteration method. A weighted factor to control the weight of new seed samples during the following training process can improve classification performance. The experiment results show that sentiment analysis of Chinese microblog based on bootstrapping not only saves much time of manual annotation but also can get better performance. The results of subjective and objective classification achieve the best accuracy rate of 62.9%, and the best accuracy rate of sentiment polarity classification is 57%.

Domain adaptation for statistical machine translation

Statistical machine translation (SMT) plays more and more important role now. The performance of the SMT is largely dependent on the size and quality of training data. But the demands for translation is rich, how to make the best of limited in-domain data to satisfy the needs of translation coming from different domains is one of the hot focus in current SMT. Domain adaption aims to obviously improve the specific-domain performance by bringing much out-of-domain parallel corpus at the absence of in-domain parallel corpus. Domain adaption is one of the keys to get the SMT into practical application. This paper introduces mainstream methods of domain adaption for SMT, compares advantages and disadvantages of representative methods based on the result of the same data and shows personal views about the possible future direction of domain adaption for SMT.

Phrase Table Combination Deficiency Analyses in Pivot-Based SMT

As the parallel corpus is not available all the time, pivot language was introduced to solve the parallel corpus sparseness in statistical machine translation. In this paper, we carried out several phrase-based SMT experiments, and analyzed the detailed reasons that caused the decline in translation performance. Experimental results indicated that both covering rate of phrase pairs and translation probability accuracy affect the quality of translation.