Definitional Question Answering Key Technologies Research

Question Answering is an important field of Information Retrieval research. Definitional Question Answering is an important branch of Question Answering, which extracts definitional sentences from open (special) document field with placed target. Both categories and properties, and description methods of concept are numerous,so there is big difference between expressions of explanatory text, then it is difficult to recognize explanatory text with current methods.Formal method of definition question answering based independent syntax mark is not conducive to express and compute explanatory semantic in text, because it doesn't involve semantic content. There is big difference in language form between different types of concepts, so recognition rate of pattern matching based all learning samples is not satisfactory. The traditional methods based on statistical analysis don't effectively calculate the fuzziness of semantic feature, which reduce the recognition rate of explanatory text. Current traditional evaluation methods base manual, some of automated methods base words, don't consider syntax and semantic features, reduce semantic accuracy.To address the issue of formal of semantic content in explanatory text, recognition rate of explanatory text, calculation of fuzziness of semantic features, and semantic feature of automated evaluation. In my research expresses explanatory text with Explanatory Content Unit (ECU), reduce the interference of independent sample with case-based reasoning, rank candidate answers with Cloud model, automated evaluate with Pyramid model.Generally, the main innovative contributions of this thesis include:â‘ Firstly a novel approach, called Case-Based Reasoning, is proposed for definitional question answering. In field of text retrieval, case-based reasoning is used for complex issue (for example: legal case retrieval). Therefore, this paper applies it to definitional question answering for the first time. Firstly, dependency grammar tree is made from explanatory text. Then, the definition case is created from grammar tree with explanatory meta-language. The approach retrieves set of cases similar with candidate answers, then rank candidate answers with these cases. BCU-CASE is created based BCU-ECU with additional phase of case retrieval. Experiment compares the average macro F value of BCU-CASE, BCU-ECU, and DT (baseline). Experimental results show that BCU-CASE 24% increase compared with DT, BCU-CASE 6% increase compared with BCU-ECU. So this approach is effective and better than DT, can improve the performance of the traditional methods based on statistical analysis.â‘¡Secondly a new algorithm based Cloud model for ranking answers is proposed for definitional question answering. The uncertainty of natural language, fuzziness and randomness in particular, has been the main issues in quantization process to be resolved. Cloud model often is used to quantify the fuzziness and randomness of natural language. Therefore, this paper applies them to ranking answers of definitional question answering for the first time. This paper creates the field of quantitative of Cloud model by ECU of definitional case, then calculate the characteristic value of cloud droplet obtained by the ECU of query case, and finally calculate the score of query case by the characteristic value. CLOUD-CASE is created based BCU-CASE exchanging ranking algorithm in additional phase of case recognition. Experiment compares the average macro F value of CLOUD-CASE, BCU-CASE, and DT (baseline). Experimental results show that CLOUD-CASE 27% increase compared with DT, CLOUD-CASE 3% increase compared with BCU-CASE. So this algorithm is effective and better than DT, the performance of the traditional methods based on statistical analysis.â‘¢Based on the Explanatory Content Unit, is proposed for formal of definitional question answering. Firstly, this paper uses dependency grammar and explanatory meta-language to parse explanatory text, and then obtains the structure of ECU. BCU-ECU is created based BCU using formal with ECU. Experiment compares the average macro F value of BCU-ECU and DT (baseline). Experimental results show that BCU-ECU 18% increase compared with DT. So this approach is effective, can improve the performance of definitional question answering.â‘£Finally a new algorithm, called Automatical Pyramid Definition Question Evaluation (APDQE), is proposed for automated evaluation of definitional question answering. In areas of the automated evaluation of summaries, Pyramid algorithm has been used as a criterion for evaluating. Therefore, this paper applies it to automated evaluation of definitional question answering for the first time. To build the pyramid model, this paper uses ECU of standard answer, then obtains the weight of ECU of test answers, and finally calculates score of standard answer. Experiments base result of APDQE, POURPRE and manual evaluation, and calculate coefficient of determination R2 between APDQE and manual evaluation, POURPRE and manual evaluation. Experimental results show that APDQE 7% increase compared with POURPRE. APDQE is better than POURPRE.