Commentary generation
One or more computing devices, systems, and/or methods for commentary generation are provided. For example, a conversation, occurring through a conversation interface associated with a content item, is monitored to identify a tone of the conversation (e.g., users discussing a news article). If the tone deviates from a target tone (e.g., a negative tone of inflammatory comments, a low participation tone, an off topic tone, etc.), then intervention is automatically and programmatically performed for the conversation. For example, subject matter of the content item, information from external sources (e.g., other articles, social network posts, or website content associated with a topic of the news article), and/or programmatically generated information (e.g., topical statements generated by a neural network) are used to construct a comment. The comment is posted to the conversation interface in order to improve the conversation, such as to increase positive engagement by users.
1. A computing device comprising:
a processor; and
memory comprising processor-executable instructions that when executed by the processor cause performance of operations, the operations comprising:
monitoring and evaluating a conversation occurring through a conversation interface associated with a content item to identify a tone of the conversation based upon comments submitted by users through the conversation interface; and
responsive to determining that the tone deviates from a target tone based upon at least one of a lack of comments or a threshold number of comments having at least one of inflammatory words or off topic statements:
identifying a subject matter of the content item for which the conversation is associated;
querying a content source to identify target content corresponding to the subject matter;
determining a plurality of snippets from the target content;
assigning weights to the plurality of snippets based upon selection criteria, wherein the assigning weights comprises at least one of:
assigning weights based upon uniqueness selection criteria, wherein the assigning weights based upon the uniqueness selection criteria comprises:
assigning a first weight to a first snippet, of the plurality of snippets, based upon a determination that the first snippet comprises one or more words matching one or more words of one or more comments within the conversation; and
assigning a second weight to a second snippet, of the plurality of snippets, based upon a determination that the second snippet does not comprise one or more words matching one or more words of one or more comments within the conversation; or
assigning weights based upon contextual selection criteria, wherein the assigning weights based upon the contextual selection criteria comprises:
assigning a third weight to a third snippet, of the plurality of snippets, based upon a determination that the third snippet comprises at least one of one or more words or one or more topics that correlate to the content item; and
assigning a fourth weight to a fourth snippet, of the plurality of snippets, based upon a determination that the fourth snippet does not comprise at least one of one or more words or one or more topics that correlate to the content item;
selecting a snippet from the plurality of snippets;
generating a context vector based upon the subject matter;
generating a natural language statement based upon the snippet and the context vector; and
posting the natural language statement as a comment through the conversation interface using at least one of a predefined name, a randomly generated name, or a determined name.
2. The computing device of claim 1 , wherein the generating the natural language statement comprises:
utilizing the snippet as a quote from the target content.
3. The computing device of claim 2 , wherein the generating the natural language statement comprises:
including the quote and a citation to the target content within the natural language statement.
4. The computing device of claim 1 , wherein the querying the content source comprises:
using the subject matter of the content item to identify structured knowledge associated with the content item.
5. The computing device of claim 4 , wherein the generating the natural language statement comprises:
generating the natural language statement to comprise an artificially generated fact derived from the structured knowledge.
6. The computing device of claim 1 , wherein the operations comprise:
determining a gender composition of one or more users of the conversation interface.
7. The computing device of claim 6 , wherein the operations comprise:
determining the determined name based upon the gender composition.
8. A non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of operations, the operations comprising:
monitoring and evaluating a conversation occurring through a conversation interface associated with a content item to identify a tone of the conversation based upon comments submitted by users through the conversation interface; and
responsive to determining that the tone deviates from a target tone based upon at least one of a lack of comments or a threshold number of comments having at least one of inflammatory words or off topic statements:
identifying a subject matter of the content item for which the conversation is associated;
querying a content source to identify target content corresponding to the subject matter;
determining a plurality of snippets from the target content;
assigning weights to the plurality of snippets based upon selection criteria, wherein the assigning weights comprises at least one of:
assigning weights based upon uniqueness selection criteria, wherein the assigning weights based upon the uniqueness selection criteria comprises:
assigning a first weight to a first snippet, of the plurality of snippets, based upon a determination that the first snippet comprises one or more words matching one or more words of one or more comments within the conversation; and
assigning a second weight to a second snippet, of the plurality of snippets, based upon a determination that the second snippet does not comprise one or more words matching one or more words of one or more comments within the conversation; or
assigning weights based upon contextual selection criteria, wherein the assigning weights based upon the contextual selection criteria comprises:
assigning a third weight to a third snippet, of the plurality of snippets, based upon a determination that the third snippet comprises at least one of one or more words or one or more topics that correlate to the content item; and
assigning a fourth weight to a fourth snippet, of the plurality of snippets, based upon a determination that the fourth snippet does not comprise at least one of one or more words or one or more topics that correlate to the content item;
selecting a snippet from the plurality of snippets;
generating a context vector based upon the subject matter;
generating a natural language statement based upon the snippet and the context vector; and
posting the natural language statement as a comment through the conversation interface using at least one of a predefined name, a randomly generated name, or a determined name.
9. The non-transitory machine readable medium of claim 8 , wherein the generating the natural language statement comprises:
utilizing the snippet as a quote from the target content.
10. The non-transitory machine readable medium of claim 9 , wherein the generating the natural language statement comprises:
including the quote and a citation to the target content within the natural language statement.
11. The non-transitory machine readable medium of claim 8 , wherein the querying the content source comprises:
using the subject matter of the content item to identify structured knowledge associated with the content item.
12. The non-transitory machine readable medium of claim 11 , wherein the generating the natural language statement comprises:
generating the natural language statement to comprise an artificially generated fact derived from the structured knowledge.
13. The non-transitory machine readable medium of claim 8 , wherein the operations comprise:
determining a gender composition of one or more users of the conversation interface.
14. The non-transitory machine readable medium of claim 13 , wherein the operations comprise:
determining the determined name based upon the gender composition.
15. A method, comprising:
monitoring and evaluating a conversation occurring through a conversation interface associated with a content item to identify a tone of the conversation based upon comments submitted by users through the conversation interface; and
responsive to determining that the tone deviates from a target tone based upon at least one of a lack of comments or a threshold number of comments having at least one of inflammatory words or off topic statements:
identifying a subject matter of the content item for which the conversation is associated;
querying a content source to identify target content corresponding to the subject matter;
determining a plurality of snippets from the target content;
assigning weights to the plurality of snippets based upon selection criteria, wherein the assigning weights comprises at least one of:
assigning weights based upon uniqueness selection criteria, wherein the assigning weights based upon the uniqueness selection criteria comprises:
assigning a first weight to a first snippet, of the plurality of snippets, based upon a determination that the first snippet comprises one or more words matching one or more words of one or more comments within the conversation; and
assigning a second weight to a second snippet, of the plurality of snippets, based upon a determination that the second snippet does not comprise one or more words matching one or more words of one or more comments within the conversation; or
assigning weights based upon contextual selection criteria, wherein the assigning weights based upon the contextual selection criteria comprises:
assigning a third weight to a third snippet, of the plurality of snippets, based upon a determination that the third snippet comprises at least one of one or more words or one or more topics that correlate to the content item; and
assigning a fourth weight to a fourth snippet, of the plurality of snippets, based upon a determination that the fourth snippet does not comprise at least one of one or more words or one or more topics that correlate to the content item;
selecting a snippet from the plurality of snippets;
generating a context vector based upon the subject matter;
generating a natural language statement based upon the snippet and the context vector; and
posting the natural language statement as a comment through the conversation interface using at least one of a predefined name, a randomly generated name, or a determined name.
16. The method of claim 15 , wherein the generating the natural language statement comprises:
utilizing the snippet as a quote from the target content.
17. The method of claim 16 , wherein the generating the natural language statement comprises:
including the quote and a citation to the target content within the natural language statement.
18. The method of claim 15 , wherein the querying the content source comprises:
using the subject matter of the content item to identify structured knowledge associated with the content item.
19. The method of claim 18 , wherein the generating the natural language statement comprises:
generating the natural language statement to comprise an artificially generated fact derived from the structured knowledge.
20. The method of claim 15 , comprising:
determining a gender composition of one or more users of the conversation interface; and
determining the determined name based upon the gender composition.