Method and apparatus for determining a topic

Patent metadata

Field	Value
Publication number	US-11366973-B2
Application number	US-201916691104-A
Country	US
Kind code	B2
Filing date	Nov 21, 2019
Priority date	Dec 28, 2018
Publication date	Jun 21, 2022
Grant date	Jun 21, 2022

Abstract

Embodiments of the present disclosure disclose a method and apparatus for determining a topic. A specific embodiment of the method comprises: determining a to-be-recognized sentence sequence; calculating similarities between the to-be-recognized sentence sequence and each of topic templates in a topic template set in a target area, the each of the topic templates in the topic template set corresponding to a topic in at least one topic in the target area, the topic template including a topic section sequence, and a topic section including a topic sentence sequence; and determining a topic of the to-be-recognized sentence sequence according to an associated parameter, the associated parameter including the similarities between the to-be-recognized sentence sequence and the each of the topic templates in the topic template set. This embodiment reduces labor costs during a topic segmentation.

First claim

What is claimed is:

1. A method for determining a topic, applied to a processor supporting a self-help dialogue system, comprising:
determining, by the processor, a to-be-recognized sentence sequence, comprising: acquiring a current dialogue sentence inputted by a user through a terminal device and a historical previous topic dialogue sentence sequence in real time, and obtaining the to-be-recognized sentence sequence by adding the current dialogue sentence to an end of the historical previous topic dialogue sentence sequence;
processing, by the processor, the to-be-recognized sentence sequence, comprising:
calculating a similarity between the to-be-recognized sentence sequence and each of topic templates in a topic template set in a target area, each of the topic templates in the topic template set corresponding to a topic in at least one topic in the target area, a topic template comprising a topic section sequence, and a topic section comprising a topic sentence sequence; and
determining a topic of the to-be-recognized sentence sequence according to an associated parameter, the associated parameter comprising the calculated similarity between the to-be-recognized sentence sequence and each of the topic templates in the topic template set; and

returning, by the processor, dialogue content of a self-help reply to the terminal device, wherein the dialogue content of the self-help reply relates to the topic of the to-be-recognized sentence sequence,
wherein the calculating the similarity between the to-be-recognized sentence sequence and each of the topic templates in the topic template set comprises executing a first similarity calculation for each of the topic templates in the topic template set, and
wherein the first similarity calculation comprises: for each of to-be-recognized sentences in the to-be-recognized sentence sequence, calculating similarities between the to-be-recognized sentence and each of topic sentences included in the topic template; determining an optimal mapping approach in at least one mapping approach by using a dynamic programming algorithm, with an aim of maximizing a similarity between the to-be-recognized sentence sequence and the topic template calculated according to a mapping approach, the mapping approach being used to correspond each of the to-be-recognized sentences in the to-be-recognized sentence sequence to a topic section in the topic template; and determining a similarity between the to-be-recognized sentence sequence and the topic template calculated according to the determined optimal mapping approach as the similarity between the to-be-recognized sentence sequence and the topic template.

2. The method according to claim 1 , wherein the determining the similarity between the to-be-recognized sentence sequence and the topic template calculated according to the determined optimal mapping approach comprises determining the similarity between the to-be-recognized sentence sequence and the topic template through a second similarity calculation, and the second similarity calculation comprises:
for each of to-be-recognized sentences in the to-be-recognized sentence sequence, determining a similarity between the to-be-recognized sentence and a mapping topic section corresponding to the to-be-recognized sentence, based on similarities between the to-be-recognized sentence and respective topic sentences in a topic sentence sequence included in the mapping topic section, wherein the mapping topic section corresponding to the to-be-recognized sentence refers to a topic section included in the topic template to which the to-be-recognized sentence is mapped according to the determined optimal mapping approach; and
determining an average value of similarities between to-be-recognized sentences in the to-be-recognized sentence sequence and the respective corresponding mapping topic sections as the similarity between the to-be-recognized sentence sequence and the topic template calculated according to the determined optimal mapping approach.

3. The method according to claim 2 , wherein the determining a similarity between the to-be-recognized sentence and a mapping topic section corresponding to the to-be-recognized sentence, based on similarities between the to-be-recognized sentence and respective topic sentences in a topic sentence sequence included in the mapping topic section comprises:
determining a maximum value in the similarities between the to-be-recognized sentence and the respective topic sentences in the topic sentence sequence included in the mapping topic section corresponding to the to-be-recognized sentence as the similarity between the to-be-recognized sentence and the corresponding mapping topic section.

4. The method according to claim 2 , wherein the determining a similarity between the to-be-recognized sentence and a mapping topic section corresponding to the to-be-recognized sentence, based on similarities between the to-be-recognized sentence and respective topic sentences in a topic sentence sequence included in the mapping topic section comprises:
determining a weighted average value of the similarities between the to-be-recognized sentence and the respective topic sentences in the topic sentence sequence included in the mapping topic section corresponding to the to-be-recognized sentence as the similarity between the to-be-recognized sentence and the corresponding mapping topic section.

5. The method according to claim 1 , wherein the determining the similarity between the to-be-recognized sentence sequence and the topic template calculated according to the determined optimal mapping approach comprises determining the similarity between the to-be-recognized sentence sequence and the topic template through a third similarity calculation, and the third similarity calculation comprises:
for each of to-be-recognized sentences in the to-be-recognized sentence sequence, determining the mapping topic section corresponding to the to-be-recognized sentence, the mapping topic section corresponding to the to-be-recognized sentence referring to the topic section included in the topic template to which the to-be-recognized sentence is mapped according to the determined optimal mapping approach;
for each of the topic sections included in the topic template, determining a similarity corresponding to the to-be-recognized sentence sequence and the topic section, based on similarities between respective topic sentences in a topic sentence sequence included in the topic section and respective to-be-recognized sentences mapped to the topic section; and
determining an average value of similarities corresponding to the to-be-recognized sentence sequence and topic sections included in the topic template as the similarity between the to-be-recognized sentence sequence and the topic template.

6. The method according to claim 1 , wherein the determining a topic of the to-be-recognized sentence sequence according to an associated parameter comprises:
determining a topic corresponding to a topic template having a maximum similarity to the to-be-recognized sentence sequence in the topic template set as the topic of the to-be-recognized sentence sequence.

7. The method according to claim 1 , wherein the associated parameter further comprises a topic and a similarity of the historical previous topic dialogue sentence sequence, the similarity of the historical previous topic dialogue sentence sequence refers to a similarity between the historical previous topic dialogue sentence sequence and a topic template corresponding to the topic of the historical previous topic dialogue sentence sequence, and the determining the topic of the to-be-recognized sentence sequence comprising:
determining a maximum value in similarities between respective topic templates

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Wang Jingwei
• Zhang Ao
• Liu Jiaxiang
• Sun Yu
• Li Zhi

Classifications

• G06F40/186 Primary

  Templates · CPC title

• G06F40/289

  Phrasal analysis, e.g. finite state techniques or chunking · CPC title

• G06F40/35 Primary

  Discourse or dialogue representation · CPC title