Web-scale entity relationship extraction that extracts pattern(s) based on an extracted tuple

1. A computer-implemented method of incremental relation extraction, comprising:

performed by one or more processors executing computer-readable instructions:

receiving a relationship seed comprising relationship data and an initial model describing an entity relationship from an input device coupled to the one or more processors;

learning a new model describing an additional entity relationship;

extracting a relation tuple comprising additional relationship data from a data corpus by applying the newly learned model;

generating one or more patterns based on the extracted relation tuple; and

selecting at least one of the one or more patterns for learning an additional new model.

2. The computer-implemented method of claim 1 iteratively performing the learning, the extracting, the generating, and the selecting.

3. The computer-implemented method of claim 2 repeating the learning, the extracting, the generating, and the selecting until no new relation tuples are extracted.

4. The computer-implemented method of claim 2 wherein the learning comprises:

learning the new model based on the relationship seed and the initial model during a first pass of the iterative process; and

learning the new model based on the one or more selected patterns during all subsequent passes of the iterative process.

5. The computer-implemented method of claim 4 wherein the extracting comprises using a logistic regression model at an entity-level.

6. The computer-implemented method of claim 4 wherein the extracting comprises using a linear-chain conditional random field model at a sentence-level.

7. The computer-implemented method of claim 4 wherein the extracting comprises using a discriminative Markov logic network (MLN) model at a page-level, a corpus-level an entity-level, and/or a sentence-level.

8. The computer-implemented method of claim 4 wherein the initial model is empty.

9. The computer-implemented method of claim 4 wherein the initial model comprises a discriminative Markov logic network (MLN) model.

10. The computer-implemented method of claim 9 wherein the received relationship seed comprises identification of a first entity and a second entity found in the data corpus.

11. The computer-implemented method of claim 10 wherein the selecting further comprises keeping a formula with a non-zero weight of an l 1 -norm regularized maximum likelihood estimation (MLE).

12. The computer-implemented method of claim 1 wherein the received relationship seed comprises identification of a first entity and a second entity found in the data corpus and one or more relation keywords.

13. The computer-implemented method of claim 1 , further comprising:

clustering the extracted relation tuples to connect same-type relation tuples; and

outputting the clustered, connected relation tuples to an output device coupled to the one or more processors.

14. The computer implemented method of claim 1 , further comprising performing open information extraction (Open IE) to identify new relationship types.

15. The computer-implemented method of claim 1 wherein the selecting comprises a structure learning problem of a Markov Logic Network (MLN) model.

16. One or more computer-readable storage devices, storing processor-executable instructions that, when executed by a processor, perform acts for incremental entity relationship extraction, the acts comprising:

iteratively mining entity relations from a data corpus using a Markov Logic Network (MLN) model comprising:

extracting entity information from the data corpus;

extracting a relation tuple from the extracted entity information based on a maximum likelihood estimation (MLE); and

generating one or more patterns based on the extracted relation tuple; and

outputting a relationship graph based on extracted relation tuples, the relationship graph having at least two entities having at least two edges, the at least two edges connected to at least two different entities.

17. The one or more computer-readable storage devices of claim 16 , wherein the entities comprise people, locations, and/or organizations.

18. The one or more computer-readable storage devices of claim 16 , wherein the data corpus comprises web documents and/or web pages available through a global and/or public network.

19. A system for implementing entity relation extraction comprising:

memory and one or more processors;

an initial seed and/or model input module, stored in the memory and executable on at least one of the one or more processors, configured to receive a relationship seed and an initial model describing an entity relationship;

a model learning module, stored in the memory and executable on at least one of the one or more processors, configured to learn a new model describing an additional entity relationship based on input from the initial seed and/or model input module or based on a pattern from a pattern selection module;

a relation tuple extraction module, stored in the memory and executable on at least one of the one or more processors, configured to extract a relation tuple from a data corpus by applying the new model;

a pattern generation module, stored in the memory and executable on at least one of the one or more processors, configured to generate one or more patterns based on the extracted relation tuples; and

the pattern selection module, stored in the memory and executable on at least one of the one or more processors, configured to assign a weight to the one or more patterns generated by the pattern generation module and select at least one of the one or more patterns based on the assigned weight to be iteratively input into the model learning module.

20. The system of claim 19 , wherein the model learning module, the relation tuple extraction module, and the pattern selection module are configured to use a discriminative MLN model, the system further comprising:

a relationship clustering module, stored in the memory and executable on at least one of the one or more processors, configured to cluster the extracted relation tuples to connect same-type relation tuples; and

an output module, stored in the memory and executable on at least one of the one or more processors, configured to output the clustered relation tuples.