Estimation of closeness of topics based on graph analytics

Patent metadata

Field	Value
Publication number	US-9542503-B2
Application number	US-201313942790-A
Country	US
Kind code	B2
Filing date	Jul 16, 2013
Priority date	Jun 11, 2013
Publication date	Jan 10, 2017
Grant date	Jan 10, 2017

Abstract

Embodiments relate to estimating closeness of topics based on graph analytics. A graph that includes a plurality of nodes and edges is accessed. Each node in the graph represents a topic and each edge represents a known association between two topics. A statistical traversal experiment is performed on the graph. A strength of relations between any two topics represented by nodes in the graph is inferred based on statistics extracted from the statistical traversal experiment.

First claim

What is claimed is:

1. A computer implemented method comprising:
accessing a graph comprised of a plurality of nodes and edges, each node representing a topic, and each edge representing a known association between two topics;
determining a first probability that, given that an agent has expressed an interest in a single topic represented by a node in the graph, the agent is interested in each of the topics represented by nodes in the graph, the determining comprising:
performing a statistical traversal experiment on said graph, the performing including using a generalized form of a matrix eigenvector algorithm that includes a Markov chain specialized to the first topic;
inferring a strength of relations between the agent and each of the topics represented by nodes in the graph, the inferring based on statistics extracted from the statistical traversal experiment; and
adjusting the inferred strength of relations to account for interests expressed in each of the topics by other agents in a reference population;
deriving a second probability that, given that the agent has expressed an interest in two or more topics represented by nodes in the graph, the agent is interested in each of the topics represented by nodes in graph; and
calculating an estimate probability by combining the first probability and the second probability using log-likelihood ratios, wherein lack of interest expressed by the agent in each of the topics is represented as subtraction using the log-likelihood ratios.

2. The method of claim 1 , further comprising constructing the graph.

3. The method of claim 1 , wherein the first topic and at least one of the other topics do not have an edge connecting them.

4. The method of claim 1 , wherein the Markov Chain has a stationary probability distribution.

5. The method of claim 1 , wherein the graph is a sparse graph.

6. The method of claim 1 , wherein the determining further comprises:
iteratively selecting a set of nodes included in the graph;
evaluating, for each iteration, a raw scoring function on the selected set; and
updating an estimate of a raw score distribution for each set included in the plurality of sets using results of the evaluation to obtain a distribution of the raw scores.

7. The method of claim 6 , wherein the raw scoring function is linear, and wherein each selected set of nodes has a single node.

8. The method of claim 6 , wherein the updated estimate of the score distribution is based on an assumption that the distribution adheres to a parametric model.

9. The method of claim 6 , further comprising:
comparing a raw score to the distribution of the raw scores;
determining a percentile of the raw score based on the comparing; and
outputting the determined percentile.

10. The method of claim 6 , wherein a raw score distribution for at least one of the sets included in the plurality of sets is pre-computed.

11. The method of claim 6 , wherein the iterative selection of the set of nodes and the evaluation of the raw scoring function on the selected set are performed as a fused operation.

Assignees

• IBM

Inventors

• Franceschini Michele M
• Jagmohan Ashish
• Lastras-Montano Luis A
• Soares Livio

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F16/9024 Primary

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06N5/046 Primary

  Forward inferencing; Production systems · CPC title

• G06F17/30958 Primary

  Physics · mapped topic

• G06N7/005

  Physics · mapped topic