Point-of-Interest Oriented Question Answering with Joint Inference of Semantic Matching and Distance Correlation

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 542–550 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 542 Point-of-Interest Oriented Question Answering with Joint Inference of Semantic Matching and Distance Correlation Yifei Yuan CUHK HongKong SAR yfyuan@se.cuhk.edu.hk Jingbo Zhou Baidu Inc. China zhoujingbo@baidu.com Wai Lam CUHK HongKong SAR wlam@se.cuhk.edu.hk Abstract Point-of-Interest (POI) oriented question an- swering (QA) aims to return a list of POIs given a question issued by a user. Recent advances in intelligent virtual assistants have opened the possibility of engaging the client software more actively in the provision of location-based services, thereby showing great promise for automatic POI retrieval. Some ex- isting QA methods can be adopted on this task such as QA similarity calculation and semantic parsing using pre-defined rules. The returned results, however, are subject to inherent limita- tions due to the lack of the ability for handling some important POI related information, in- cluding tags, location entities, and proximity- related terms (e.g. “nearby”, “close”). In this paper, we present a novel deep learning frame- work integrated with joint inference to cap- ture both tag semantic and geographic corre- lation between question and POIs. One char- acteristic of our model is to propose a special cross attention question embedding neural net- work structure to obtain question-to-POI and POI-to-question information. Besides, we uti- lize a skewed distribution to simulate the spa- tial relationship between questions and POIs. By measuring the results offered by the model against existing methods, we demonstrate its robustness and practicability, and supplement our conclusions with empirical evidence. 1 Introduction Point-of-Interest (POI) oriented question answer- ing (QA) problem is a special QA task which aims to answer users’ questions by generating a list of POIs. With the rapid development of smart agents (e.g. Amazon Echo) and intelligent virtual assis- tants (e.g. Apple Siri and Google Assistant), there are many POI oriented queries being requested everyday. Some examples of these questions are “Where can I take my kid to have fun nearby New York City” or “Where can we go in LA with my friends”. The answers are typically a list of POIs such as parks, malls, or restaurants corresponding to the details provided by users in the questions. According to some statistics, there are millions of POI oriented QA questions being requested per day on a mobile search engine in China. Generally speaking, semantic parsing and simi- larity matching methods are utilized to tackle the POI oriented QA problem in current solutions. Nev- ertheless, both of them are subject to inherent lim- itations and deserve to be improved. Semantic parsing based methods convert the questions to for- mal representations (such as SQL queries) using pre-defined rules, then get the POI results from the query. Difficulties arise, however, when the form of the questions varies from person to person. Be- sides, due to ambiguous expressions of a specific tag, semantic parsing methods always fail …

Content truncated to 500 words