System and methods for quantitative assessment of information in natural language contents and for determining relevance using association data

• US 9,201,927 B1
• Filed: 01/01/2013
• Issued: 12/01/2015
• Est. Priority Date: 01/07/2009
• Status: Active Grant

• Current Assignee: Linfo IP LLC (Pueblo Nuevo LLC)
• Original Assignee: Guangsheng Zhang
• Inventors: Zhang, Guangsheng
• Primary Examiner(s): HO, BINH VAN
• Application Number:
  US13/732,374
• Time in Patent Office: 1,064 Days
• Field of Search: 707/730
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

System and methods are disclosed for quantitatively assessing information in natural language contents related to an object name, or a concept or topic name, and for determining the relevance between a text content and one or more concepts or topics. The methods include identifying the grammatical or semantic attributes and other contextual information of terms in the text content, and retrieving an object-specific dataset related to the object name, or an equivalent to a concept or a topic name. The data set includes property names and association-strength values. The methods further include matching the terms in the text content with the property terms in the dataset, and calculating a score as a quantitative measure of the relevance between the text content and the concept or topic.

Claims

1. A method implemented on a computer comprising a processor, and for determining relevance between a text content and an object or a topic, the method comprising:

receiving a text content comprising one or more words or phrases or sentences as terms, and tokenizing the text content into one or more tokens, each being an instance of a term in the text content;

identifying a grammatical attribute, or a semantic attribute, or an external term frequency associated with the one or more tokens or terms in the text content, wherein the grammatical attribute includes at least a subject, a predicate or part of a predicate, a modifier in a phrase, a head of a phrase, a sub-phrase of a phrase, an object, a noun, a verb, an adjective, or an adverb, wherein the semantic attribute includes at least semantic roles and attribute values, wherein the external term frequency is obtained from text contents other than the received text content;

determining an importance measure for each token or term based on the grammatical attribute, or the semantic attribute, or the external term frequency;

receiving one or more datasets, wherein each dataset is associated with a name or description representing an object, wherein the object comprises a physical or conceptual object, a topic, or a pre-defined attribute, and wherein each dataset comprises one or more words or phrases as names of properties associated with the corresponding object, wherein the names of properties represent other objects or concepts or topics or attributes-related to the object, wherein the names of properties collectively represent a type of definition or representation of the object;

matching at least two tokens or terms in the text content with at least two property names in each of the one or more datasets;

for each of the one or more datasets, producing a score based at least on the importance measure of the token or term that matches a property name in the dataset, when the importance measure is in the form of a term importance score that is calculated based on the external frequency, or based on the grammatical attribute, or based on the semantic attribute or attribute value, and when the score based on the importance measure is in the form of a relevance score, the relevance score is produced as a function of the term importance score; and

marking or selecting one or more of the names or descriptions representing the one or more objects as being relevant to the text content if the corresponding score is above a predefined threshold.

2. The method of claim 1, further comprising:

outputting, for storage or display, the one or more selected names or descriptions, or the score associated with the one or more marked or selected names or descriptions, wherein the function of the score includes at least serving as a quantitative measure of relevance between the text content and the object.

3. The method of claim 1, wherein a weighting co-efficient is assigned to each token based on the grammatical attribute or semantic attribute or attribute value associated with the token or term in the text content, wherein the relevance score is further produced as a function of at least the weighting co-efficient assigned to the token that matches a property name.

4. The method of claim 1, wherein a property name in each of the datasets has a corresponding numerical value representing the association strength between the property name and the object, wherein the score is produced further based on the association strength value associated with the matched property name.

5. The method of claim 1, wherein each property name in the dataset has a corresponding numerical value representing the association strength between the property name and the object, wherein the text content is a sentence comprising at least a first token and a second token, and the first token is associated with the grammatical role of a subject and the second token is associated with a grammatical role of at least part of a predicate;

or wherein the text content is a multi-word phrase comprising at least a first token and a second token, and the first token is associated with the grammatical role of a head of the multi-word phrase and the second token is associated with a grammatical role of a modifier in the multi-word phrase, or the first token and the second token are associated with a grammatical role of a sub-phrase in the multi-word phrase;

wherein the score is produced further based on the association strength value associated with the property name that matches the first token and the association strength value associated with the property name that matches the second token.

6. The method of claim 1, wherein a property name in each of the datasets has more than one corresponding numerical values representing the association strength between the property name and the object, each value representing the association strength under a specific context, wherein the step of producing the score is further based on the association strength value associated with the matched property name under a specific context, wherein the context includes at least the grammatical or semantic attribute of the matching tokens or terms.

7. The method of claim 1, wherein the property names in the datasets are divided into a plurality of groups, wherein the score is produced further based on the group or groups that comprise the property names matching the token.

8. The method of claim 1, wherein each property name in the dataset has a corresponding numerical value representing the association strength between the property name and the concept or topic or attribute, wherein the property names are divided into groups based on their association strength values being above or below a predetermined threshold, wherein the score is produced further based on the group or groups that comprise the property names matching the token.

9. The method of claim 1, wherein the one or more datasets are represented in the form of a table, each dataset representing one or more columns or rows in the table;

or

wherein the one or more datasets are represented in the form of an inverted index with each term in the index pointing to a corresponding object name with an optional corresponding numerical value, wherein the function of the numerical value includes representing the association strength or association status between the term and the object;

or

wherein the one or more datasets are represented in the form of a hierarchical structure with one or more property terms as first-level property terms each associated with one or more second-level property terms or with corresponding second-level association strength values;

or

wherein the one or more datasets are represented in the form of the one or more property terms being stored separately as data points linked with the object name or description.

10. The method of claim 1, wherein the one or more marked or selected names or descriptions represent one or more document categories, the method further comprising:

assigning the text content to the one or more document categories represented by the one or more marked or selected names or descriptions.

11. The method of claim 1, wherein the one or more selected names or descriptions represent the context of a word or phrase in the text content, wherein the word or phrase has a plurality of meanings or senses, the method further comprising:

selecting a meaning or sense from the plurality of meanings or senses as a relevant meaning or sense of the word or phrase to the context represented by the one or more marked or selected names or descriptions.

12. The method of claim 1, wherein the one or more marked or selected names or descriptions represent one or more topics or topic domains, the method further comprising:

outputting one or more of the marked or selected names or descriptions as one or more answers to a question about related topics or topic domains of the text content.

13. The method of claim 1, wherein the one or more marked or selected names or descriptions represent one or more topics or topic domains relevant for advertisements, the method further comprising:

indicating the one or more marked or selected names or descriptions for use with advertising, or displaying an advertisement in a user interface in connection with the text content, wherein the user interface includes a web user interface, a social network page, a chat screen, or an email interface, or as an audio message through an audio device, on a computing or communication device including a desktop or a notebook computer, a mobile or hand-held device, a cloud-based device, or a mobile or land phone.

14. The method of claim 1, wherein the text content is a word or a multi-word phrase, or a sentence or a paragraph or a document or a collection of a plurality of documents, wherein a document or a collection of a plurality of documents is any content containing text of a language, including a search query, a webpage, a blog, an email, an SMS message, a transcript from an audio or text conversation, a user comment or chat message entered on a website or on a social network page, or in a user interface of a computing or communication device, wherein the computing or communication device includes a desktop computer, a notebook computer, a mobile or hand-held device, a cloud-based device, or a mobile phone or a land phone.

15. A computer system for identifying one or more topics or topic domains or attributes related to a text content, comprising:

a computer processor operable to

receive a text content comprising one or more words or phrases or sentences as terms, and to tokenize the text content into one or more tokens, each being an instance of a term in the text content;

identify a grammatical attribute, or a semantic attribute, or an external term frequency associated with the one or more tokens or terms in the text content, wherein the grammatical attribute includes at least a subject, a predicate or part of a predicate, a modifier in a phrase, a head of a phrase, a sub-phrase of a phrase, an object, a noun, a verb, an adjective, or an adverb, wherein the semantic attribute includes at least semantic roles and attribute values, wherein the external term frequency is obtained from text contents other than the received text content;

determine an importance measure for each token or term based on the grammatical attribute, or the semantic attribute, or the external term frequency;

receive one or more datasets, wherein each dataset is associated with a name or description representing an object, wherein the object comprises a physical object or a concept, a topic, a topic domain, or a pre-defined attribute, wherein each dataset comprises one or more words or phrases as names of properties associated with the corresponding object, wherein the names of properties represent other objects or concepts or topics or attributes related to the object, wherein the names of properties collectively represent a type of definition or representation of the object;

match at least two tokens or terms in the text content with at least two property names in each of the one or more datasets;

produce, for each of the one or more datasets, a score based at least on the importance measure of each token or term that matches a property name in the dataset, when the importance measure is in the form of a term importance score that is calculated based on the external frequency, or based on the grammatical attribute, or based on the semantic attribute or attribute value, and when the score based on the importance measure is in the form of a relevance score, the relevance score is produced as a function of the term importance score;

select one or more names or descriptions of the one or more datasets as relevant objects or topics or topic domains or attributes to the text content if the corresponding score is above a predefined threshold; and

display, or provide an instruction to display in a user interface, or to store in the computer storage, the selected one or more names or descriptions, or the score associated with the one or more selected names or descriptions, wherein the function of the score includes serving as a quantitative measure of relevance between the text content and the selected one or more names or descriptions of the objects, or topics or topic domains or attributes.

16. The computer system of claim 15, wherein the text content is a search query, a webpage, a blog, an email, an SMS message, a transcript from an audio or text conversation, a user comment or chat message entered on a website or on a social network page, or in a user interface of a computing or communication device, wherein the selected topic or domain is relevant for advertisements, wherein the computer processor is further operable to:

display in a user interface, or provide an instruction to display, an advertisement in connection with the text content, wherein the user interface includes a web user interface, a social network page, a chat screen, or an email interface, or as an audio message through an audio device, on a computing or communication device including a desktop or a notebook computer, a mobile or hand-held device, a cloud-based device, or a mobile or land phone.

17. The computer system of claim 15, wherein the text content further comprises a first content and a second content, wherein the selected one or more objects or topics or domains or attributes are relevant to both the first and the second content, wherein the first content or the second content includes a resume or a job description, wherein the computer processor is further operable to:

display or store, or provide an instruction to display or store, the first content as a relevant content to the second content, or the second content as a relevant content to the first content.

18. The computer system of claim 15, wherein the text content is a document with text, wherein the selected one or more names or descriptions represent one or more document categories or classes including at least the class or category of spam or junk email and non-spam email, wherein the computer processor is further operable to:

classify or categorize the document into the one or more classes or categories.

19. A computer system for searching documents and ranking search results based on association, comprising:

a computer storage storing one or more datasets, each dataset having a name or description representing an object, wherein the object comprises a physical object or a topic or concept, and each dataset comprises one or more words or phrases as names of properties associated with the corresponding named object, wherein the names of properties represent other objects or concepts or topics or attributes related to the object, wherein the names of properties collectively represent a type of definition or representation of the object; and

a computer processor operable to

receive a search query,

receive one or more documents each comprising one or more words or phrases or sentences as terms,

tokenize each of the documents into one or more tokens, each being an instance of a term in the text content,

identify a grammatical attribute, or a semantic attribute, or an external term frequency associated with the one or more tokens or terms in the text content, wherein the grammatical attribute includes at least a subject, a predicate or part of a predicate, a modifier in a phrase, a head of a phrase, a sub-phrase of a phrase, an object, a noun, a verb, an adjective, or an adverb, wherein the semantic attribute includes at least semantic roles and attribute values, wherein the external term frequency is obtained from text contents other than the received documents,

determine an importance measure for each token or term based on the grammatical attribute, or the semantic attribute, or the external term frequency,

receive one or more names or descriptions of the datasets, wherein the one or more names or descriptions match a word or phrase in the query,

obtain one or more of the corresponding datasets,

for at least one of the documents, match at least two tokens or terms in the document with at least two property names in at least one of the one or more datasets,

for at least one of the one or more datasets or documents, produce a relevance score based at least on the importance measure of the token or term that matches a property name in the dataset, when the importance measure is in the form of a term importance score that is calculated based on the external frequency, or based on the grammatical attribute, or based on the semantic attribute or attribute value, the relevance score is produced as a function of the term importance score,

select one or more of the documents as relevant documents to the query and rank the selected documents based at least on the relevance score, and

output the ranked documents or document representations as a search result.