Methods for searching with semantic similarity scores in one or more ontologies

• US 20110040766A1
• Filed: 08/13/2009
• Published: 02/17/2011
• Est. Priority Date: 08/13/2009
• Status: Active Grant

• Current Assignee: Charite-Universitatsmedizin Berlin, Max-Planck-Gesellschaft Zur Foerderung Der Wissenschaften e.V. (Max-Planck-Gesellschaft zur Frderung der Wissenschaften e)
• Original Assignee: Charite-Universitatsmedizin Berlin, Max-Planck-Gesellschaft Zur Foerderung Der Wissenschaften e.V. (Max-Planck-Gesellschaft zur Frderung der Wissenschaften e)
• Inventors: Robinson, Peter N., Köhler, Sebastian, Bauer, Sebastian, Schulz, Marcel H.
• Granted Patent: US 9,002,857 B2
• Time in Patent Office: Days
• US Class Current: 707/749
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/38 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually

Abstract

A method assigns importance ranks to documents within repositories or databases, such as any database of documents such as books or other printed material, electronic documentation, and pages within the world-wide web. The method uses a corpus of indexed documents that has been annotated to the terms of one or more ontologies in order to assign a semantic similarity score to queries based on terms taken from the ontologies. A statistical model is used to test the significance of matches between query terms and documents or categories. The method results in an acceleration of over 10,000-fold for realistic queries and ontologies, and makes it practicable to calculate P-values dynamically or to keep database annotations and the related P-value distributions up to date by frequent recalculation.

Claims

1. A method of scoring items comprising

obtaining a plurality of items having annotations that are semantically related to the terms in a query by using one or more ontologies and said annotations in order to determine the statistical significance of similarities between query terms and sets of terms used in the annotations of the items,

assigning a P-value to the probability of obtaining the observed semantic similarity score for each obtained item, and

scoring the items according to their individual P-values.

4. The method according to claim 1, wherein the items are ranked according to their P-value, with the most relevant query results being those items assigned the lowest P-values.

6. The method of claim 1, whereby collections of documents or books are processed to determine the set of ontology terms that best annotate them by text-mining of the back cover text, the table of contents, or the entire text.

7. The method of claim 1, wherein

for organizations that have experts that are responsible for certain categories, the most relevant ontology term or terms are assigned to each expert,

for each new request for help, the method of claim 1 is used to find the expert most suitable for processing the request or answering a question, and

a signal identifying the most suitable expert is generated.

8. The method of claim 1, wherein in order to offer wares for sale in certain categories one or more ontologies are used to annotate the wares with the ontology terms.

9. The method of claim 1 wherein said ontologies are used as vocabularies to describe symptoms or diseases of patients, and wherein a search for candidate diagnoses is carried out in a database based on said ontologies.

10. The method of claim 1 wherein said ontologies are used as vocabularies to describe drugs and chemical compounds, and wherein a search for candidate drugs and chemical compounds is carried out in a database based on said ontologies.

11. The method of claim 1 wherein said ontologies are used as vocabularies to describe a user'"'"'s personal characteristics, e.g. preferences, appearance, wishes, dreams, and life plans, and wherein a search for other users having matching personal characteristics is carried out in a database based on said ontologies.

12. The method of claim 1, wherein said ontologies are used to describe attributes of departments and/or individual workers and wherein a search for a department or worker is carried out in a database based on said ontologies.

13. The method of claim 1, wherein said ontologies are used to describe attributes of jobs and individuals and wherein a search for matches between jobs and individuals is carried out in a database based on said ontologies.

14. The method of claim 1, wherein said ontologies are used as vocabularies to describe the properties of services, e.g. travel products.

15. The method of claim 1, wherein said ontologies are used to process websites in the world-wide web and/or documents that are made accessible via the world-wide web.

16. The method of claim 1, wherein

the P-values are compared to an adjustable P-value cutoff, and

objects are returned whose P-values pass the adjustable P-value cutoff.

17. The method of claim 1, whereby a heuristic is used to decide which ontologies to analyze.

18. The method of claim 1, whereby parallel processing computational techniques are used to analyze several ontologies in parallel in order to process a query with the method of claim 1.

19. The method of claim 1, whereby said ontologies are used to formulate queries in the semantic web.

20. The method of claim 1, wherein multiple ontologies are used to annotate items s from an online or offline classified advertising service that offers users multiple classifications or categories in which to offer items for sale.

2. A method of scoring items comprising the steps of:

obtaining a plurality of items that are semantically related to the terms in a query by using one or more ontologies and annotations to them in order to determine the statistical significance of similarity scores between query terms and sets of terms used to annotate the items,

collapsing portions of said ontologies that contribute equally to final score by identifying ontology terms, which contribute equally to the score, and by calculating their contribution to the score distribution, and

using the calculated contribution of said collapsed portions to determine the statistical significance of similarities between said query terms and said sets of terms used to annotate the items.

5. The method of claim 2, whereby the distribution of similarity scores is weighted by the frequency of term annotations in a domain.

3. A method of scoring items comprising the steps of:

obtaining a plurality of items that are semantically related to the terms in a query by using one or more attribute ontologies and annotations to them in order to determine the similarities between query terms and sets of terms used to annotate the items, and

using ontological similarity score measures derived from attribute ontologies to rank said items according to one or more ontologies.