Method and device for input prediction

Patent metadata

Field	Value
Publication number	US-11409374-B2
Application number	US-201916761216-A
Country	US
Kind code	B2
Filing date	Jun 18, 2019
Priority date	Jun 28, 2018
Publication date	Aug 9, 2022
Grant date	Aug 9, 2022

Abstract

A method and apparatus for input prediction. The method includes: obtaining an input current text; obtaining a first state parameter of a first text from a cache; inputting the current text and the first state parameter into the recurrent neural network, determining a state parameter of the current text through the recurrent neural network according to the first state parameter; and determining a predicted text of the current text from the word library according to the state parameter of the current text. The first text is a previous text of the current text; the first state parameter is determined through a preset recurrent neural network according to the first text and a state parameter of a previous text of the first text; the recurrent neural network is trained with a preset word library; the word library is used to store words. Through the method and apparatus, time for determining a predicted text is reduced.

First claim

What is claimed is:

1. A method for input prediction, comprising:
obtaining an input current text;
obtaining a first state parameter of a first text from a cache, wherein the first text is a previous text of the current text; the first state parameter is determined through a preset recurrent neural network according to the first text and a state parameter of a previous text of the first text; the recurrent neural network is trained with a word library, wherein the word library is preset; the word library is used to store words, wherein a state parameter of a text can reflect association between the text and words in the word library;
inputting the current text and the first state parameter into the recurrent neural network, and determining a state parameter of the current text through the recurrent neural network according to the first state parameter; and
determining a predicted text of the current text from the word library according to the state parameter of the current text.

2. The method according to claim 1 , wherein, after determining the state parameter of the current text, the method further comprises:
storing the state parameter of the current text in the cache.

3. The method according to claim 1 , wherein, determining a predicted text of the current text from the word library according to the state parameter of the current text comprises:
determining word scores for words in the word library according to the state parameter of the current text;
selecting target words from the words in the word library according to the word scores; and
determining a predicted text of the current text according to the target words.

4. The method according to claim 3 , wherein, determining a predicted text of the current text according to the target words comprises:
matching the current text with morphemes in a preset dictionary library respectively, and taking morphemes in the dictionary library that match the current text as candidate morphemes that are similar to the current text, wherein, the dictionary library is used to store morphemes;
obtaining word scores for words in the word library determined in the input prediction of the first text;
determining scores for the candidate morphemes from the word scores for words in the word library determined in the input prediction of the first text; and
selecting the predicted text of the current text from the target words and candidate morphemes according to word scores for the target words and the scores for the candidate morphemes.

5. The method according to claim 1 , wherein, determining the state parameter of the current text through a recurrent neural network according to the first state parameter comprises:
determining the state parameter of the current text according to network parameters for the operation of the recurrent neural network and the first state parameter;
wherein, the network parameters for the operation of the recurrent neural network are obtained in the following manner:
obtaining network parameters when training of the recurrent neural network is completed; and
performing integer approximation on fractional parameters in the network parameters when the training is completed, and taking network parameters subjected to the integer approximation as the network parameters for the operation of the recurrent neural network.

6. The method according to claim 1 , wherein, the method is applied to a client, and an installation file of the client is obtained by the following operations:
obtaining an initial installation file generated according to original codes of the client, and obtaining an operation function in the original codes of the client; and
removing, from the initial installation file, operation functions other than the operation function in the original codes to obtain the installation file of the client.

7. An electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;
the memory is configured for storing a computer program; and
the processor is configured for executing the computer program stored in the memory to perform the input prediction method according to claim 1 .

8. A non-transitory computer-readable storage medium having a computer program stored thereon which, when executed by a processor, performs the input prediction method according to claim 1 .

9. An apparatus for input prediction, comprising a memory that stores executable modules; and a processor, coupled to the memory, that executes the executable modules, the executable modules comprising:
a text obtaining module configured for obtaining an input current text;
a parameter obtaining module configured for obtaining a first state parameter of a first text from a cache, wherein the first text is a previous text of the current text; the first state parameter is determined through a preset recurrent neural network according to the first text and a state parameter of a previous text of the first text; the recurrent neural network is trained with a word library, wherein the word library is preset; the word library is used to store words, wherein a state parameter of a text can reflect association between the text and words in the word library;
a parameter determining module configured for inputting the current text and the first state parameter into the recurrent neural network, and determining a state parameter of the current text through the recurrent neural network according to the first state parameter; and
a text predicting module configured for determining a predicted text of the current text from the word library according to the state parameter of the current text.

10. The apparatus according to claim 9 , further comprising:
a parameter caching module, configured for storing the state parameter of the current text in the cache after determining the state parameter of the current text.

11. The apparatus according to claim 9 , wherein, the text predicting module comprises:
a first determining sub-module configured for determining word scores for words in the word library according to the state parameter of the current text;
a selecting sub-module configured for selecting target words from the words in the word library according to the word scores; and
a second determining sub-module, configured for determining a predicted text of the current text according to the target words.

12. The apparatus according to claim 11 , wherein, the second determining sub-module is further configured for:
matching the current text with morphemes in a preset dictionary library respectively, and taking morphemes in the dictionary library that match the current text as candidate morphemes that are similar to the current text, wherein, the dictionary library is used to store morphemes;
obtaining word scores for words in the word library determined in the input prediction of the first text;
determining scores for the candidate morphemes from the word scores for words in the word library determined in the input prediction of the first text; and
selecting the predicted text of the current text from the target words and candidate morphemes according to word scores for the target words and the scores for the candidate morphemes.

13. The apparatus according to claim 9 , wherein, determining the state parameter of the current text through a recurrent neural network according to the first state parameter comprises:
determining the state parameter of the current text according to network parameters for an operation of the recurrent neural network a

Assignees

• Beijing Kingsoft Internet Security Software Co Ltd

Inventors

• Jia Yawei
• Wu Xiaoqiang

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title