Answer Extraction System Based on Latent Dirichlet Allocation

Question Answering (QA) task is still an active area of research in information retrieval. A variety of methods which have been proposed in the literature during the last few decades to solve this task have achieved mixed success. However, such methods developed in the Arabic language are scarce and do not have a good performance record. This is due to the challenges of Arabic language. QA based on Frequently Asked Questions is an important branch of QA in which a question is answered based on pre-answered ones. In this paper, the aim is to build a question answering system that responds to a user inquiry based on pre-answered questions. The proposed approach is based on Latent Dirichlet Allocation. Firstly, the dataset, pairs of questions and associated answers, will be grouped into several clusters of related documents. Next, when a new question to be answered is posed to the system, it,therefore, starts to assign this question to its appropriate cluster, then, use a similarity measure to get the top ten closest possible answers. Preliminary results show that the proposed method is achieving a good level of performance.

@article{Ali2016,
  title     = {Answer Extraction System Based on Latent Dirichlet Allocation},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {7},
  number    = {4},
  year      = {2016},
  publisher = {The Science and Information Organization},
  author    = {Mohammed A. S. Ali and Sherif M. Abdou},
  doi       = {10.14569/IJACSA.2016.070461},
  url       = {https://doi.org/10.14569/IJACSA.2016.070461}
}