Context-based metadata generation and automatic annotation of electronic media in a computer network

1. A method for improving search and retrieval of input items being published to a computer network, the method comprising:

while publishing an input item to the computer network:

generating, by a content generation module executing at a computing device and based at least in part on information about the input item, a descriptive context of the input item by at least:

performing a query for the input item by searching an academic database that includes at least one of: books, research papers, or journal articles or searching the Internet using a publicly available search engine; and

adding, to the descriptive context of the input item, at least one textual item from a response to the query for the input item;

comparing, by a comparison module executing at the computing device, the descriptive context of the input item to respective descriptive contexts of a plurality of other items previously published to the computer network to determine, using a text similarity model, respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, wherein each of the plurality of other items is already tagged with one or more existing tags that a search engine searches when performing a query on the computer network;

automatically tagging, by the content generation module, and based at least in part on the respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, the input item with at least one existing tag from the one or more existing tags that a particular content item from the plurality of content items is tagged; and

after automatically tagging the input item, automatically publishing, by the computing device, the item to the computer network.

2. The method of claim 1 , further comprising:

generating, by the computing device, the respective descriptive contexts of each the plurality of other items.

3. The method of claim 1 , wherein comparing the descriptive context of the input item to the respective descriptive contexts of each of the plurality of other items to determine the respective levels of similarity between the input item and each of the plurality of other items comprises:

generating, based at least in part on the descriptive context of the input item, a vector representing a semantic makeup of the descriptive context of the input item; and

determining respective cosine similarities between the vector representing the semantic makeup of the descriptive context of the input item and each of a plurality of respective vectors representing respective semantic makeups of the respective descriptive contexts of the plurality of other items.

4. The method of claim 3 , wherein the vector representing the semantic makeup of the descriptive context of the input item comprises at least one of: a Term Frequency-Inverse Document Frequency (TF-IDF) vector, a Latent Dirichlet Allocation TF-IDF (LDA-TFIDF) vector, or a Latent Semantic Indexing TF-IDF (LSI-TFIDF) vector.

5. The method of claim 1 , wherein the plurality of other items comprises a plurality of topics, and wherein tagging the input item with at least one existing tag from the one or more existing tags that the particular content item from the plurality of content items is tagged comprises tagging the input item with at least one topic from the plurality of topics.

6. The method of claim 1 , wherein the plurality of other items comprises a plurality of topics, the method further comprising generating the plurality of topics by:

performing a topic query for each potential topic from a set of potential topics using the potential topic; and

adding, to the plurality of topics, at least one item from a respective response to the topic query for the potential topic.

7. The method of claim 6 , further comprising obtaining the set of potential topics from a database of research requests.

8. The method of claim 6 , wherein performing the respective query for the potential topic comprises searching a crowd-sourced information site using the potential topic.

9. The method of claim 1 , wherein the plurality of other content items comprises a subset of content items, the method further comprising:

determining, as part of the subset of content items, each content item from the plurality of other content items that is tagged with at least one existing tag with which the input item is also tagged;

generating, based at least in part on the one or more existing tags with which each of the subset of content items is tagged, a set of candidate tags; and

ranking, based at least in part on the respective levels of similarity between the input item and each of the subset of content items, the set of candidate tags to form a ranked set of candidate tags,

wherein tagging the input item is based at least in part on the ranked set of candidate tags.

10. The method of claim 9 , further comprising:

clustering the set of candidate tags in a semantic space; and

pruning, based at least on the clustering, the set of candidate tags by retaining only candidate tags that are contained within a particular cluster, the particular cluster containing existing tags of the input item.

11. The method of claim 10 , wherein clustering the set of candidate tags comprises clustering the set of candidate tags using a web-distance metric.

12. The method of claim 1 , wherein the input item comprises a research paper.

13. The method of claim 1 , wherein the information about the input item comprises a document title.

14. The method of claim 1 , wherein the information about the input item comprises an existing tag with which the input item is tagged.

15. The method of claim 1 , further comprising:

generating, by the computing device, a topic database for use in annotating input items, the topic database specifying a plurality of topics;

constructing, by the computing device, a global descriptive context for each topic in the topic database; and

ranking, by the computing device, each topic in the topic database by comparing the respective descriptive context for each of the topics to the descriptive context for the input item.

16. The method of claim 15 , wherein constructing a global descriptive context comprises:

accessing a plurality of external data sources to retrieve potential topics;

aggregating the topics into a topic database;

conducting a topic query in the publicly available search engine for each topic in the topic database;

constructing the global descriptive context based at least in part on the results of the topic query.

17. The method of claim 15 , wherein performing the normalization process includes conducting a topic query for the one or more topics in the topic database using a crowd-sourced information source, and updating the topic database based at least in part on the results of the topic query.

18. A method for improving search and retrieval of input items being published to a computer network, the method comprising:

while publishing one or more input items to the computer network:

identifying, by a computing device, one or more external data sources;

retrieving, by the computing device, topics from the identified one or more external data sources;

aggregating, by the computing device, the topics into a topic database;

generalizing, by the computing device, the topic database by performing a normalization process;

conducting a query, by the computing device in a publicly available search engine, for each topic in the generalized topic database;

constructing, by the computing device and based at least in part on the results of the query, a topic context database;

identifying, by the computing device, information about the one or more input items, wherein the information about the input item comprises a title;

conducting a query, by the computing device in a publicly available search engine, for content in the title for the one or more input items;

constructing, by the computing device and based at least in part on the results of the query, a respective title context for the one or more input items;

comparing, by the computing device, the title context and the one or more topics in the topic context database using a text similarity computation model;

determine, by the computing device and based at least in part on the results of the comparison, one or more topics from the topic context database with which to annotate the one or more input items; and

after annotating, by the computing device, the one or more input items based on the determined topics, publishing, by the computing device, the one or more items to the computer network.

19. The method of claim 18 , wherein performing the normalization process includes conducting a query for the one or more topics in the topic database using a crowd-sourced information source, and updating the topic database based at least in part on the results of the search query.

20. A computing device having a processor configured to:

while publishing an input item to a computer network:

generate, by executing a content generation module, based at least in part on information about an input item, a descriptive context of the input item by at least:

performing a query for the input item by searching an academic database that includes at least one of: books, research papers, or journal articles or searching the Internet using a publicly available search engine; and

adding, to the descriptive context of the input item, at least one textual item from a response to the query for the input item;

compare, by executing a comparison module, the descriptive context of the input item to respective descriptive contexts of a plurality of other items previously published to the computer network to determine, using a text similarity model, respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, wherein each of the plurality of other items is already tagged with one or more existing tags that a search engine searches when performing a query on the computer network; and

automatically tag, using the content generation module, and based at least in part on the respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, the input item with at least one existing tag from the one or more existing tags that a particular content item from the plurality of content items is tagged; and

after automatically tagging the input item with the at least one existing tag, automatically publish the input item to the computer network.

21. The computing device of claim 20 , wherein comparing the descriptive context of the input item to the respective descriptive contexts of the plurality of other items to determine the respective levels of similarity between the input item and each of the plurality of other items comprises:

generating, based at least in part on the descriptive context of the input item, a vector representing a semantic makeup of the descriptive context of the input item; and

determining respective cosine similarities between the vector representing the semantic makeup of the descriptive context of the input item and each of a plurality of respective vectors representing respective semantic makeups of the respective descriptive contexts of the plurality of other items.

22. The computing device of claim 20 further configured to:

generate a topic database for use in annotating input items, the topic database specifying a plurality of topics;

construct a global context for each topic in the topic database; and

rank each topic in the topic database by comparing the respective descriptive context for each of the topics to the descriptive context for the input item.

23. A computer-readable storage medium encoded with instructions that, when executed, cause at least one processor to:

while publishing an input item to a computer network:

generate, based at least in part on information about the input item, a descriptive context of the input item by at least:

performing a query for the input item by searching an academic database that includes at least one of: books, research papers, or journal articles or searching the Internet using a publicly available search engine; and

adding, to the descriptive context of the input item at least one textual item from a response to the query for the input item;

compare the descriptive context of the input item to respective descriptive contexts of a plurality of other items previously published to the computer network to determine, using a text similarity model, respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, wherein each of the plurality of other items is already tagged with one or more existing tags that a search engine searches when performing a query on the computer network; and

tag based at least in part on the respective levels of similarity between the descriptive context of the input item and the respective descriptive context of each of the plurality of other items, the input item with at least one existing tag from the one or more existing tags that a particular content item from the plurality of content items is tagged; and

tagging the input item with the at least one existing tag, publish the input item to the computer network.