Automatically generating conference minutes

1. A method, comprising:

acquiring a text conference record through speech recognition on audio data;

dividing the text conference record into a plurality of conference paragraphs, generating a conference paragraph summary for each conference paragraph of the plurality of conference paragraphs, and generating, by using a neural network, a conference record summary based on the conference paragraph summary of each conference paragraph;

extracting conference instructions based on the text conference record; and

generating conference minutes based on the conference record summary and the conference instructions, wherein

the conference instructions comprise an entity instruction, an opinion instruction, and an attitude instruction, and

the extracting the conference instructions based on the text conference record comprises extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record, wherein the extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record comprises:

performing lexical analysis on each sentence in the text conference record to identify participant entities and topic entities of the conference;

extracting opinion features from the text conference record to identify a plurality of sentences that represent opinions, and extracting opinion instructions corresponding to the participant entities by reorganizing the plurality of sentences; and

extracting attitude features from the text conference record to identify a plurality of sentences that represent attitudes, and extracting attitude instructions corresponding to the participant entities by reorganizing the plurality of sentences,

wherein the method further comprising a statistics collection act,

wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances of each participant entity or a proportion of utterances of each participant entity in total utterances; and

determining utterance enthusiasm of each participant entity based on the one or more of the number of utterances or the proportion of utterances of the participant entity,

or, wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances on each topic entity or a proportion of utterances on each topic entity in the total utterances; and

determining utterance enthusiasm on each topic entity based on the one or more of the number of utterances on the topic entity or the proportion of utterances on the topic entity in the total utterances.

2. The method of claim 1 , wherein the generating the conference paragraph summary for each conference paragraph comprises:

evaluating each sentence in the conference paragraph to obtain an evaluation value of the sentence;

determining candidate sentences based on the evaluation value of each sentence in the conference paragraph to form a candidate sentence set; and

generating the conference paragraph summary based on the candidate sentence set.

3. The method of claim 1 , wherein the acquiring the text conference record comprises:

acquiring a conference recording as the audio data; and

performing speech recognition on the conference recording to form the text conference record.

4. The method of claim 2 , wherein the evaluating each sentence in the conference paragraph to obtain the evaluation value of the sentence comprises:

calculating a score value, a redundancy value, and a coherence value of the sentence; and

calculating the evaluation value based on the score value, the redundancy value, and the coherence value.

5. The method of claim 4 , wherein

the score value is calculated through a neural network, and a higher score value indicates that the sentence contains more concerned information.

6. The method of claim 4 , wherein

the redundancy value is obtained by analyzing a similarity between the sentence and other sentences in the paragraph, and a higher redundancy value indicates that a degree of repetition of the sentence is greater.

7. The method of claim 4 , wherein

the coherence value is obtained by analyzing a correlation between the sentence and other sentences in the paragraph, and a higher coherence value indicates that a relationship between the sentence and the other sentences in the paragraph is closer.

8. The method of claim 4 , wherein the determining the candidate sentences based on the evaluation value of each sentence in the conference paragraph to form the candidate sentence set comprises:

selecting sentences that meet predetermined conditions to form the candidate sentence set,

wherein the predetermined conditions comprise: the evaluation value being greater than an evaluation value threshold, the score value being greater than a score value threshold, the redundancy value being greater than a redundancy value threshold, and the coherence value being greater than a coherence value threshold.

9. The method of claim 1 , comprising one or more of:

extracting opinions and attitudes of different participant entities on a same topic entity based on the entity instruction, the opinion instruction, and the attitude instruction; and

extracting opinions and attitudes of a same participant entity on different topic entities based on the entity instruction, the opinion instruction, and the attitude instruction.

10. The method of claim 1 , comprising:

identifying synonyms of the conference instructions through word vector calculation; and

normalizing the synonyms to identify description for a same conference instruction.

11. The method of claim 1 , wherein

the attitude instruction comprises attitude tendencies, and the attitude tendencies comprise positive, negative, and neutral attitude tendencies.

12. The method of claim 1 , wherein the generating the conference record summary based on the conference paragraph summary of each conference paragraph comprises:

evaluating each summary sentence in the conference paragraph summary to obtain an evaluation value of the summary sentence;

determining candidate summary sentences based on the evaluation value of each summary sentence in the conference paragraph summary to form a candidate summary sentence set; and

generating the conference record summary based on the candidate summary sentence set.

13. The method of claim 12 , wherein the evaluating each summary sentence in the conference paragraph summary to obtain the evaluation value of the summary sentence comprises:

calculating a score value, a redundancy value, and a coherence value of the summary sentence; and

calculating the evaluation value of the summary sentence based on the score value, the redundancy value, and the coherence value of the summary sentence.

14. The method of claim 1 , further comprising an automatic extraction act,

wherein the automatic extraction act includes:

extracting keywords and high-frequency words from the text conference record by performing semantic calculation on the text conference record and grading each word in the text conference record; and

determining at least one of conference background, conference topics, conference objectives, conference conclusions, and task arrangements based on the keywords and the high-frequency words.

15. An electronic device, comprising:

a processor; and

a memory that stores a program, the program comprising instructions that, when executed by the processor, cause the processor to perform acts including:

acquiring a text conference record through speech recognition on audio data;

dividing the text conference record into a plurality of conference paragraphs, generating a conference paragraph summary for each conference paragraph of the plurality of conference paragraphs, and generating, by using a neural network, a conference record summary based on the conference paragraph summary of each conference paragraph;

extracting conference instructions based on the text conference record; and

generating conference minutes based on the conference record summary and the conference instructions, wherein

the conference instructions comprise an entity instruction, an opinion instruction, and an attitude instruction, and

the extracting the conference instructions based on the text conference record comprises extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record, wherein the extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record comprises:

performing lexical analysis on each sentence in the text conference record to identify participant entities and topic entities of the conference;

extracting opinion features from the text conference record to identify a plurality of sentences that represent opinions, and extracting opinion instructions corresponding to the participant entities by reorganizing the plurality of sentences; and

extracting attitude features from the text conference record to identify a plurality of sentences that represent attitudes, and extracting attitude instructions corresponding to the participant entities by reorganizing the plurality of sentences,

wherein the processor is further caused to perform a statistics collection act,

wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances of each participant entity or a proportion of utterances of each participant entity in total utterances; and

determining utterance enthusiasm of each participant entity based on the one or more of the number of utterances or the proportion of utterances of the participant entity,

or, wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances on each topic entity or a proportion of utterances on each topic entity in the total utterances; and

determining utterance enthusiasm on each topic entity based on the one or more of the number of utterances on the topic entity or the proportion of utterances on the topic entity in the total utterances.

16. A non-transitory computer-readable storage medium that stores a program, the program comprising instructions that, when executed by a processor of an electronic device, cause the electronic device to perform acts including:

acquiring a text conference record through speech recognition on audio data;

dividing the text conference record into a plurality of conference paragraphs, generating a conference paragraph summary for each conference paragraph of the plurality of conference paragraphs, and generating, by using a neural network, a conference record summary based on the conference paragraph summary of each conference paragraph;

extracting conference instructions based on the text conference record; and

generating conference minutes based on the conference record summary and the conference instructions, wherein

the conference instructions comprise an entity instruction, an opinion instruction, and an attitude instruction, and

the extracting the conference instructions based on the text conference record comprises extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record, wherein the extracting the entity instruction, the opinion instruction, and the attitude instruction from the text conference record comprises:

performing lexical analysis on each sentence in the text conference record to identify participant entities and topic entities of the conference;

extracting opinion features from the text conference record to identify a plurality of sentences that represent opinions, and extracting opinion instructions corresponding to the participant entities by reorganizing the plurality of sentences; and

extracting attitude features from the text conference record to identify a plurality of sentences that represent attitudes, and extracting attitude instructions corresponding to the participant entities by reorganizing the plurality of sentences,

wherein the electronic device is further caused to perform a statistics collection act,

wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances of each participant entity or a proportion of utterances of each participant entity in total utterances; and

determining utterance enthusiasm of each participant entity based on the one or more of the number of utterances or the proportion of utterances of the participant entity,

or, wherein the statistics collection act includes:

collecting statistics about one or more of a number of utterances on each topic entity or a proportion of utterances on each topic entity in the total utterances; and

determining utterance enthusiasm on each topic entity based on the one or more of the number of utterances on the topic entity or the proportion of utterances on the topic entity in the total utterances.