Back to results
US 10,504,145 B2Granted
Automated classification of network-accessible content based on events
AssigneeGumGum, Inc.
InventorRoopal Garg
Life of this patent
Estimating the term…
Abstract
Systems and methods are provided for generating and using classification models to automatically classify pages or other source content as including text about one or more real-world events. Generating the classification models may include analyzing text content of a large number of different pages from both a reference source and from more dynamic sources, such as from publisher sources via a network. Features for training classifiers may be determined based in part on the top n-grams identified among pages that have been determined to be associated with a given event.
Claims20
A computer system comprising:
- memory;
- and
- a computing device, comprising a physical processor, that is in communication with the memory and that is configured with processor-executable instructions to perform operations comprising:
- for each event of a plurality of events:
- retrieve a reference page associated with the event, wherein the reference page comprises narrative text regarding the event, wherein the reference page further comprises a plurality of references to other pages;
- access the other pages referenced in the reference page;
- generate a first set of terms associated with the event, wherein each of the terms in the first set comprises one or more words, wherein the first set includes (a) one or more terms appearing in text of the reference page and (b) one or terms appearing in text of at least one of the other pages referenced in the reference page;
- determine a plurality of network-accessible pages regarding the event based at least in part on determinations that each page of the plurality of network-accessible pages includes a name of the event within a uniform resource identifier for the page;
- generate a second set of terms associated with the event, wherein each of the terms of the second set comprises one or more words appearing together in at least one of the plurality of network-accessible pages regarding the event;
- for each term in the first set of terms and second set of terms associated with the event, generate a score for the term that represents a strength of association between the term and the event, wherein the score for each term is generated based at least in part on a number of times that the term appears in pages associated with the event relative to (a) a first frequency with which the term appears in pages associated with other events and (b) a second frequency with which the term is used in an entire collection of content;
- select a plurality of top scoring terms associated with the event, wherein the plurality of top scoring terms are selected from among the first set of terms and the second set of terms;
- and
- generate one or more classification models for determining whether an input page includes text content regarding one or more of the plurality of events, wherein at least a subset of the top scoring terms associated with each event are used as features in training the one or more classification models.
10 more elements
A computer-implemented method comprising:
- as implemented by one or more computing devices configured with specific executable instructions,
- retrieving a name of an event;
- accessing a reference page associated with the event, wherein the reference page comprises narrative text regarding the event, wherein the reference page further comprises a plurality of references to other pages;
- accessing the other pages referenced in the reference page;
- generating a first set of terms associated with the event, wherein each of the terms in the first set comprises one or more words, wherein the first set includes (a) one or more terms appearing in text of the reference page and (b) one or terms appearing in text of at least one of the other pages referenced in the reference page;
- determining a plurality of network-accessible pages regarding the event based at least in part on determinations that each page of the plurality of network-accessible pages includes a name of the event within a uniform resource identifier for the page;
- generating a second set of terms associated with the event, wherein each of the terms of the second set comprises one or more words appearing together in at least one of the plurality of network-accessible pages regarding the event;
- for each term in the first set of terms and second set of terms associated with the event, generating a score for the term that represents a strength of association between the term and the event, wherein the score for each term is generated based at least in part on a number of times that the term appears in pages associated with the event relative to at least a frequency with which the term appears in pages associated with other events;
- selecting a plurality of top scoring terms associated with the event, wherein the plurality of top scoring terms are selected from among the first set of terms and the second set of terms;
- and
- generating a classification model for determining whether an input page includes text content regarding the event, wherein at least a subset of the top scoring terms associated with the event are used as features in training the classification model.
8 more elements
Dependencies are inferred from the claim text.