Query-Bag-Matching-CIKM

This repo contains the composed Quora dataset for the short paper "Query-bag Matching]{Query-bag Matching with Mutual Coverage for Information-seeking Conversations in E-commerce" in CIKM 2019. It also contains the detailed experiments of this paper.

An Example of Alime

• An example of the query-bag pair and the corresponding answer.
• User query:
• 怎么电话小二? (How to call you?)
• Bag (Pre-defined questions):
• 怎么联系你们? (How to contact you?)
• 客服热线是多少? (What's the hotline?)
• Answer:
• 电话是123456 (The phone number is 123456.)

Composed Quora Dataset

• The dataset is in the dataset directory.
• The Quora dataset is composed from the Quora Question Pairs Dataset

Experiment Setup

Data

• max sequence length: 20

hCNN

Convolution2d for cross-attention matrix

• Conv1: kernel size 6 ; num output 8 ; stride 1 ; paddding "SAME" ; activation Relu ;
• Max pooling1: pool size 4,4 ; stride 4 ; padding "SAME"
• Conv2: kernel size 4 ; num output 16 ; stride 3 ; paddding "SAME" ; activation Relu ;
• Max pooling2: pool size 2,2 ; stride 2 ; padding "SAME"

Convolution1d for query and question

• filters 100 ; kernel size 2/3/4 ; max pooling pool size 20-kernel_size + 1 ; max pooling stride 1