Facet Annotation by Extending CNN with a Matching Strategy

Area of Science:

• Information Science
• Computer Science
• Artificial Intelligence

Background:

• Community Question Answering (CQA) websites organize vast amounts of question-answer pairs (QAPs).
• Current topic-based organizations often fail to meet users' need for fine-grained search. Topic facets offer a solution for navigating and refining QAP collections.
• Facet annotation can improve the organization and retrieval of information within CQA platforms.

Purpose of the Study:

• To propose and evaluate FACM, a model for automatic facet annotation of QAPs.
• To enhance the searchability and organization of QAPs in CQA systems.
• To address the challenge of facet heterogeneity in large-scale QAP datasets.

Main Methods:

• Incorporating phrase information into text representation using Convolution Neural Networks (CNNs) with varying kernel sizes.
• Employing a matching strategy between QAPs and facet label texts (FaLTs) sourced from Wikipedia to generate similarity matrices.
• Training a three-channel CNN for the precise assignment of facets to QAPs.

Main Results:

• The proposed FACM model demonstrates superior performance compared to existing state-of-the-art methods.
• Experiments conducted on three real-world datasets validate the effectiveness of FACM in facet label assignment.
• The model successfully handles facet heterogeneity, improving the accuracy of QAP organization.

Conclusions:

• FACM offers an effective approach for annotating QAPs with facets, significantly improving information retrieval.
• The integration of CNNs and a matching strategy provides a robust solution for fine-grained search in CQA systems.
• This work advances the field of information organization and retrieval in large-scale question-answering platforms.