Classifying relevance of natural language text for topic-based notifications

A server-system for generating topic-relevant notifications based on topics of interest detected within content items by natural language processing, comprising:

• one or more processors programmed with computer program instructions that, when executed, cause the server-system:
• identify, from a plurality of sources for respective pluralities of content items, initial content item information for the content items;
• apply, based on a source and an identified content item, at least one filter to initial content item information to determine whether the identified content item exceeds a threshold likelihood of being potentially relevant to one or more of a plurality of topics of interest;
• obtain, in response to determining the content item is potentially relevant to at least one of the plurality of topics of interest, additional content item information for the content item from the source of the content item based on at least some of the initial content item information;
• construct a content item record comprising natural language texts parsed from the initial and the additional content item information for a plurality of different information sub-types, a given portion of natural language text being parsed from content item information for a given one of the information sub-types based on detection of one or more corresponding sub-type identifiers within the content item information;
• provide the content item record to a natural language processing model, wherein the natural language processing model is configured to:
• identify a topic interest for the content item record based on distances between chunks of natural language text within the content item record to chunks of natural language text indicative of at least one respective topic of interest, and
• score the identified topic of interest based on at least some of the distances, the at least some distances corresponding to chunks of natural language text indicative of at least the identified topic of interest;
• detect, in association with at least one information sub-type populated for the content item record, a chunk of natural language text associated with value indicative of a score of the detected chunk of natural language text relative to other possible chunks of natural language text having respectively associated values for the information sub-type;
• generate, in response to the score of the identified topic of interest exceeding a threshold indicative of content-topic relevance and the score of the detected chunk of natural language text exceeding a threshold indicative of impact on the identified topic of interest, a notification indicative of the content item and the identified topic of interest;
• and
• transmit the notification to a plurality of client devices based on received selections of the users of the client devices being indicative of interest in the identified topic of interest.