Content recommendations
View Patent ↗Users consume a wide variety of content from various sources, such as videos accessible through websites. As provided herein, content recommendations that are contextually and/or semantically relevant to current content consumed by a user may be identified and provided to the user. For example, metadata for a video being watched by the user may be identified (e.g., terms extracted from a description, user reviews, a category, and/or other information). The metadata may be used to identify content recommendations based upon the metadata corresponding to terms grouped into a set of refined topic groupings of a graph comprising terms and relationships between terms extracted from a content corpus. The metadata may be matched to relevant terms within the set of refined topic groupings, and content recommendations comprising content corresponding to the relevant terms may be suggested to the user.
1. A method for providing content recommendations, comprising:
generating a graph based upon terms extracted from a content corpus, the graph comprising one or more nodes connected by one or more edges, a first edge connecting a first node and a second node, the first node representing a first term, the second node representing a second term, and the first edge representing a relationship between the first term and the second term;
decomposing the graph into a set of topic groupings, a topic grouping comprising one or more terms that contribute to a topic of the topic grouping;
filtering the set of topic groupings based upon a topic contribution threshold, specifying a threshold amount of contribution a term is to provide to an associated topic, to identify a set of refined topic groupings from the graph,
the filtering comprising:
assigning weights to respective terms within the topic grouping;
normalizing the weights to define a probability mass function;
computing an entropy of the probability mass function;
responsive to the entropy having a value that is not between a minimum entropy curve and a uniform probability distribution entropy curve, removing a third term from the topic grouping based upon the topic contribution threshold; and
removing a fourth term from the topic grouping responsive to determining that the fourth term provides a first amount of contribution, less than the threshold amount of contribution, to the topic;
identifying metadata of current content consumed by a user;
projecting the metadata into the set of refined topic groupings within the graph to identify a set of relevant nodes; and
providing one or more content recommendations based upon content associated with the set of relevant nodes.
2. The method of claim 1 , the filtering the set of topic groupings comprising:
performing a multiscale harmonic analysis of the graph to obtain a set of scaling functions, a scaling function corresponding to a topic of a refined topic grouping.
3. The method of claim 2 , the projecting comprising:
extracting a term from the metadata; and
projecting the term into the set of refined topic groupings using the set of scaling functions to identify a relevant node.
4. The method of claim 2 , the projecting comprising:
projecting the metadata into a Euclidean space, within which the set of refined topic groupings is represented, using the set of scaling functions.
5. The method of claim 4 , the projecting comprising:
identifying a threshold number of neighboring nodes within the Euclidean space, for inclusion within the set of relevant nodes, utilizing a locality hashing technique.
6. The method of claim 1 , the identifying metadata comprising:
identifying at least one of a non-entity term, an adjective term, an adverb term, a noun term, a verb term, an entity term, or a term from the metadata.
7. The method of claim 1 , the current content comprising a first video, and a content recommendation comprising a second video contextually relevant to the first video.
8. The method of claim 1 , the current content associated with a first language, and a content recommendation associated with a second language.
9. The method of claim 1 , the current content associated with a non-English language, and a content recommendation associated with the non-English language.
10. The method of claim 1 , the filtering comprising:
utilizing a diffusion wavelet tree construction technique to identify the set of refined topic groupings.
11. The method of claim 10 , the utilizing a diffusion wavelet tree construction technique comprising:
removing one or more global supportive terms from the topic grouping; and
retaining one or more local supportive terms within the topic grouping.
12. The method of claim 1 , the filtering the set of topic groupings comprising:
iteratively removing one or more terms from the topic grouping based upon the topic contribution threshold.
13. The method of claim 1 , the minimum entropy curve corresponding to the set of topic groupings and the uniform probability distribution entropy curve corresponding to the set of topic groupings.
14. The method of claim 1 , the removing a third term comprising:
reducing the entropy based upon a removed term.
15. A system for providing content recommendations, comprising:
a graph generation component configured to:
generate a graph based upon terms extracted from a content corpus;
identify a set of topic groupings within the graph, a topic grouping comprising one or more terms that contribute to a topic of the topic grouping; and
filter the set of topic groupings based upon a topic contribution threshold, specifying a threshold amount of contribution a term is to provide to an associated topic, to identify a set of refined topic groupings within the graph, the filtering comprising:
removing a first term from the topic grouping responsive to:
determining that the first term provides a first amount of contribution to the topic, wherein the first amount of contribution corresponds to a first amount of semantic relevance of the first term to the topic; and
determining that the first amount of contribution is less than the threshold amount of contribution, wherein the threshold amount of contribution corresponds to a threshold amount of semantic relevance;
assigning weights to respective terms within the topic grouping;
normalizing the weights to define a probability mass function;
computing an entropy of the probability mass function; and
responsive to the entropy having a value that is not between a minimum entropy curve and a uniform probability distribution entropy curve, removing a second term from the topic grouping based upon the topic contribution threshold; and
a content recommendation component configured to:
identify metadata of current content consumed by a user;
evaluate the set of refined topic groupings from the graph based upon the metadata to identify a set of relevant nodes; and
provide one or more content recommendations based upon content associated with the set of relevant nodes.
16. The system of claim 15 , the graph generation component configured to:
iteratively remove one or more terms from the topic grouping based upon the topic contribution threshold.
17. The system of claim 15 , the minimum entropy curve corresponding to the set of topic groupings and the uniform probability distribution entropy curve corresponding to the set of topic groupings.
18. The system of claim 15 , the graph generation component configured to:
utilize a diffusion wavelet tree construction technique to identify the set of refined topic groupings.
19. The system of claim 18 , the graph generation component configured to:
utilize the diffusion wavelet tree construction technique to:
remove one or more global supportive terms from the topic grouping; and
retain one or more local supportive terms within the topic grouping.
20. A nontransitory memory device storing instructions that, when executed on a processor of a server, cause the server to provide content recommendations, by:
decomposing a graph, comprising one or more terms extracted from a content corpus, into a set of topic groupings, a topic grouping comprising one or more terms that contribute to a topic of the topic grouping;
filtering the set of topic groupings based upon a topic contribution threshold, specifying a threshold amount of contribution a term is to provide to an associated topic, to identify a set of refined topic groupings from the graph, the filtering comprising:
assigning weights to respective terms within the topic grouping;
normalizing the weights to define a probability mass function;
computing an entropy of the probability mass function; and
responsive to the entropy having a value that is not between a minimum entropy curve and a uniform probability distribution entropy curve, removing a second term from the topic grouping based upon the topic contribution threshold;
identifying metadata of a current content consumed by a user;
evaluating the set of refined topic groupings within the graph based upon the metadata to identify a set of relevant nodes; and
providing one or more content recommendations based upon content associated with the set of relevant nodes.