Method and apparatus for pushing information

Patent metadata

Field	Value
Publication number	US-11023505-B2
Application number	US-201815926584-A
Country	US
Kind code	B2
Filing date	Mar 20, 2018
Priority date	Jun 29, 2017
Publication date	Jun 1, 2021
Grant date	Jun 1, 2021

Abstract

A method and apparatus for pushing information. A specific implementation of the method includes: parsing page content browsed by a user to extract a keyword of the page content and determine a theme corresponding to the keyword; determining a preset keyword matching the keyword in a preset keyword collection; determining an associated keyword associated with the determined preset keyword based on a pre-established associated relationship of the preset keyword; and pushing information corresponding to the theme and/or the associated keyword to the user. This implementation implements targeted information push.

First claim

What is claimed is:

1. A method for pushing information, the method comprising:
parsing page content browsed by a user to extract a keyword of the page content and determine a theme corresponding to the keyword;
determining a preset keyword matching the keyword in a preset keyword collection;
determining an associated keyword associated with the determined preset keyword based on a pre-established associated relationship of the preset keyword; and
pushing information corresponding to the theme or the associated keyword to the user,
wherein the parsing page content browsed by a user to extract a keyword of the page content and determine a theme corresponding to the keyword comprises:
parsing the page content browsed by the user to extract the keyword of the page content, and inputting the keyword into a pre-generated theme probability distribution model to obtain at least one first candidate theme and a probability of determining each one of the at least one first candidate theme as the theme corresponding to the keyword;
inputting the keyword into a pre-generated generalization model to obtain at least one second candidate theme and a similarity between the keyword and each one of the at least one second candidate theme; and
fusing the at least one first candidate theme and the at least one second candidate theme to determine the theme corresponding to the keyword, and
wherein, fusing the at least one first candidate theme and the at least one second candidate theme to determine the theme corresponding to the keyword comprises:
determining one or more same themes among the at least one first candidate theme and the at least one second candidate theme;
determining a value obtained by performing a weighted summation on the probability and similarity corresponding to each of the same themes according to a preset weight as a first probability;
determining a similarity corresponding to each second candidate theme that is different from the first candidate theme as a second probability; and
selecting a preset number of themes from the same themes and the second candidate theme that is different from the first candidate theme according to the first and second probabilities as the theme corresponding to the keyword.

2. The method according to claim 1 , wherein before the parsing page content browsed by the user to extract the keyword of the page content and determine the theme corresponding to the keyword, the method further comprises generating the theme probability distribution model, comprising:
parsing a plurality of preset texts to determine a keyword and a theme of each text and establish a corresponding relationship between the keyword and the theme of the text;
performing statistical calculation on the established corresponding relationship to determine a number of times of establishing the corresponding relationship between each keyword and each theme; and
generating the theme probability distribution model corresponding to the each determined keyword based on the number of times of establishing the corresponding relationship between the each keyword and the each theme.

3. The method according to claim 1 , wherein before the parsing page content browsed by the user to extract the keyword of the page content and determine the theme corresponding to the keyword, the method further comprises generating the generalization model, comprising:
parsing a plurality of preset texts to generate a keyword collection for each text and determine a theme of the each text;
merging the keyword collections corresponding to the texts having same themes to generate a keyword collection corresponding to the each determined theme;
for the each determined theme, extracting a characteristic keyword from the keyword collection corresponding to the theme based on a chi-square test method; and
training by using the characteristic keyword corresponding to the each theme as an input based on a machine learning method to obtain the generalization model.

4. The method according to claim 3 , wherein the parsing a plurality of preset texts to generate a keyword collection for each text and determine a theme of the each text comprises:
performing, for the each text among the plurality of preset texts, a word segmentation on the text, and deleting a stop word, a preposition and an adverb among a plurality of words obtained after the word segmentation to obtain the keyword collection corresponding to the text; and
inputting the each text into a pretrained theme model to determine the theme of the each text, the theme model being configured to characterize a corresponding relationship between the text and the theme.

5. The method according to claim 1 , wherein before the parsing page content browsed by a user to extract a keyword of the page content and determine a theme corresponding to the keyword, the method further comprises establishing the associated relationship of the preset keyword, comprising:
parsing historical search data and historical browsing data to determine a first keyword and at least one second keyword associated with each first keyword;
respectively calculating a similarity between the first keyword and each one of the at least one second keyword, and each preset keyword in the preset keyword collection, to determine a preset keyword in the preset keyword collection having a maximum similarity to the first keyword as a first preset keyword, and respectively determine a preset keyword in the preset keyword collection having a maximum similarity to each second keyword as a second preset keyword; and
respectively establishing the associated relationship between the first preset keyword and each second preset keyword.

6. The method according to claim 5 , wherein after the respectively establishing the associated relationship between the first preset keyword and each second preset keyword, the method further comprises:
counting up a number of co-occurrences of the first preset keyword and the each second preset keyword according to the historical search data and the historical browsing data; and
determining a probability of transition from the first preset keyword to the each second preset keyword based on the determined number of co-occurrences, the probability of transition from the first preset keyword to the each second preset keyword being a ratio of the number of co-occurrences of the first preset keyword and the second preset keyword to a sum of the each determined number of co-occurrences.

7. The method according to claim 6 , wherein the determining an associated keyword associated with the determined preset keyword based on a pre-established associated relationship of the preset keyword comprises:
determining the determined preset keyword as a target first preset keyword, determining a second preset keyword associated with the target first preset keyword according to the established associated relationship, and extracting the probability of transition from the target first preset keyword to the each associated second preset keyword; and
determining the second preset keyword having a probability of transition greater than a preset probability as the associated keyword associated with the target first preset keyword.

8. The method according to claim 1 , wherein after the parsing page content browsed by a user to extract a keyword of the page content and determine a theme corresponding to the keyword, the method further comprises:
presenting a link of the page content on a web page corresponding to the theme.

9. An apparatus for pushing information, the apparatus comprising:
at least one processor; and
a memory storing instructions, which when executed by the at least one processor, cause the at least one processor

Assignees

• Beijing Baidu Netcom Sci & Tec

Inventors

• Chi Jianfeng
• Guo Weidong
• Lv Yajuan
• Chen Yile
• Luo Jinwen

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/048

  Activation functions · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title