Semantic Role Labeling (SRL) as a Shallow Semantic Parsing causes more and more attention recently. The shortage of manually tagged data is one of main obstacles to supervised learning, which is even serious in SRL. Transductive SVM (TSVM) is a novel semi-supervised learning method special to small mount of tagged data. In this paper, we introduce an application of TSVM in Chinese SRL. To improve the performance of TSVM, some heuristics have been designed from the semantic perspective. The experiment results on Chinese Propbank showed that TSVM outperforms SVM in small tagged data, and after using heuristics, it performs further better.

Semantic parsing is a fundamental problem in natural language understanding. Shallow semantic parsing, as methodology, can be a novel strategy. the rationale and development of shallow semantic parsing is explored, and the definitions, corpus and approaches to semantic role labeling are introduced. A comparison is made between two semantic role systems in English and one in Chinese, and features of statistic methods are surveyed in a syntactic and semantic view.

Various semantic web service discovery techniques have been proposed, many of which perform the profile based service signature (I/O) matching. However, the service I/O concepts are not sufficient to discover web services accurately. This paper presents a new method to enhance the semantic description of semantic web service by using the semantic constraints of service I/O concepts in specific context. The semantic constraints described in a constraint graph are extracted automatically from the parsing results of the service description text by a set of heuristic rules. The corresponding semantic web service matchmaker performs not only the profile’s semantic matching but also the matching of their semantic constraints with the help of a constraint graph based matchmaking algorithm. The experiment results are encouraging when applying the semantic constraint to discover semantic web services on the service retrieval test collection OWLS-TC v2.

Semantic analysis is one of the fundamental and key problems in the research of content-based Text Mining.Most of supervised machine learning methods led to poor performance when work on limited tagged data.This paper investigated a novel semi-supervised learning algorithm—Transductive Support Vector Machine for shallow semantic parsing.An optimization strategy of selecting training instances,based on active learning,was integrated with TSVM.The experiment result shows that the method integrating TSVM and optimization strategy for shallow semantic parsing outperforms supervised methods on small tagged data.