Latent Space Embedding for Retrieval in Question-Answer Archives

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 855–865 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Latent Space Embedding for Retrieval in Question-Answer Archives Deepak P1 Dinesh Garg2 Shirish Shevade3 1Queen’s University Belfast, Northern Ireland, UK deepaksp@acm.org 2Indian Institute of Technology Gandhinagar, India dgarg@iitgn.ac.in 3Indian Institute of Science, Bangalore, India shirish@csa.iisc.ernet.in Abstract Community-driven Question Answering (CQA) systems such as Yahoo! Answers have become valuable sources of reusable information. CQA retrieval enables usage of historical CQA archives to solve new questions posed by users. This task has re- ceived much recent attention, with meth- ods building upon literature from transla- tion models, topic models, and deep learn- ing. In this paper, we devise a CQA re- trieval technique, LASER-QA, that embeds question-answer pairs within a unified la- tent space preserving the local neighbor- hood structure of question and answer spaces. The idea is that such a space mirrors semantic similarity among ques- tions as well as answers, thereby enabling high quality retrieval. Through an em- pirical analysis on various real-world QA datasets, we illustrate the improved effec- tiveness of LASER-QA over state-of-the- art methods. 1 Introduction Community-based Question Answering (CQA) services such as Yahoo! Answers1, Quora2, Stack- Overflow3, and Baidu Zhidao4 have become a de- pendable source of knowledge to solve common user problems. These allow a user to post queries such as how and why questions that seek descrip- tive solutions and opinions as answers. Over time, these services build up a large archive of question- answer knowledge that may be leveraged to solve new user questions. The CQA retrieval problem, 1https://answers.yahoo.com/ 2https://www.quora.com/ 3http://stackoverflow.com/ 4https://en.wikipedia.org/wiki/Baidu Knows Table 1: Example CQA Pairs # QA Cause 1 Q: My internet connection is not working, my router shows the ”Internet” led blinking in red. Router A: Please go to the router login page and re-login Authentication with broadband credentials; click ”connect” and Issue you should be on the internet. 2 Q: My internet connection is not working, only the power led is lit in the router. Router A: Can you check whether the broadband cable Loose is plugged in. Maybe, the broadband cable is not Connection connected properly. that has received much recent attention, is about addressing this opportunity. CQA retrieval meth- ods focus on finding historical archived knowl- edge (questions, answers or QA pairs) that are rel- evant to a newly posed user question. The cen- tral technical challenge that differentiates CQA re- trieval from other general purpose IR tasks is that of the need to address the lexical gap (aka lexi- cal chasm) in QA archives. Lexical chasm means that text fragments in questions (e.g., disk full) may lead to semantically correlated content in an- swers (e.g., format). This QA-correlation is differ- ent from semantic relatedness such as synonymy and antonymy; in the above example, the correla- tion is due to disk full issues often leading to so- lution involving disk formating. Explicit correla- tion modelling, using statistical translation mod- els, have met …

Content truncated to 500 words