Entity Topic Models for Mining Documents

Entity Topic Models for Mining Documents
H/T Bill Slawski 
A new way of discovering documents and text associated with a business will change how search finds relevancy to a business.
These associations will be powerful for relating the various documents related to business and will go far beyond keywords to concepts.

Ping David Amerland 

#seofornow   #entitytopicmodels   #contentstrategy  

Originally shared by Bill Slawski

ETM: Entity Topic Models for Mining Documents Associated with Entities

This is an interesting look at different topic models that might often be described as well organized when they may also be shown as being related to organizations, places, people, and other pages on the Web.

Here are some lines from the document:

To deal with the different types of attributes associated with documents, different topic models have been proposed: (1) Topic Over Time [18] and Dynamic Topic Models [3] are designed for documents with timestamps, (2) GeoFolk [13] and Latent Geographical Topic Analysis [19] are proposed for documents with GPS information, (3) Author Models [9] and Autor Topic Models [12] deal with documents with author lists, and (4) Link-LDA [6] and Block-LDA [1] are designed for dealing with documents with hyperlinks, citations, and other forms of link information.

The ability to capture the association of documents with real-world entities or concepts holds great promise over traditional keyword-based approaches (cf. Google’s “knowledge graph”, which enhances search results by linking documents to entities1 ). In a similar vein, we argue that it is also highly desirable to build topic models that can capture the complex patterns involving the entities associated with documents. Almost any document is associated with some set of real-worlds entities.

#TopicModels #EntityModels #KnowledgeGraph  

http://hanj.cs.illinois.edu/pdf/icdm12_hkim.pdf