Method and apparatus for collecting entity aliases

1. A method for collecting entity aliases, the method comprising:

obtaining at least one web object in a result produced by a web search engine in response to a web search query based upon an input entity;

obtaining, from a social tag website in response to obtaining the at least one web object, a respective set of social tags that is associated, on the social tag web site, with the at least one web objects, the respective set of social tags comprising at least one alias for at least one web object within the at least one web objects;

defining, based upon a relationship on the social tag website between social tags within the respective set of social tags and the web objects, candidate aliases for the input entity, the candidate aliases comprising values of social tags within the respective set of social tags;

extracting features from the candidate aliases, the extracting being based upon at least one of:

a word within a candidate alias within the candidate aliases;

at least one of a word preceding and following the candidate alias within the candidate aliases;

capitalization of a beginning letter of a word within the candidate aliases;

a prefix of a word within the candidate alias within the candidate aliases;

a postfix of a word within a candidate alias within the candidate aliases; and

a semantic type of a word within a candidate alias within the candidate aliases;

detecting, using a processor, a set of detected entities within the candidate aliases by applying a named entity recognition engine to the candidate aliases, the each detected entity within the set of detected entities comprising a respective feature vector set representing features extracted from the candidate aliases, the applying the named entity recognition engine comprising:

creating a learned model based upon feature vector sets from a first plurality of detected entities; and

applying the learned model to a second plurality of detected entities, the second plurality of detected entities being different than the first plurality;

ranking each of the detected entities according to a respective quality criteria comprising any one of or a combination of any of:

the homogeneity of the detected entity with the input entity;

the occurrence frequency of the tags which contain the detected entity; and

the authoritativeness of the web objects from which the entity is obtained;

collecting detected entities having a rank higher than a predetermined threshold to obtain a collection of aliases of the input entities; and

storing the collection of aliases of the input entities into a dataset.

2. An apparatus for collecting entity aliases, comprising:

a web object collection component adapted to obtain web objects in a result produced by a web search engine in response to a web search query based upon an input entity;

a social tag collection component adapted to:

obtain, from a social tag website in response to the web object collection component obtaining the web objects, a respective set of social tags that are associated, on the social tag web site, with the web objects; and

define, based upon a relationship on the social tag website between social tags within the respective set of social tags and the web objects, candidate aliases for the input entity, the candidate aliases comprising values of social tags within the respective set of social tags; and

an aliases refinement processor adapted to refine the candidate aliases using a process comprising:

extracting features from the candidate aliases, the extracting being based upon at least one of:

a word within a candidate alias within the candidate aliases;

at least one of a word preceding and following the candidate alias within the candidate aliases;

capitalization of a beginning letter of a word within the candidate aliases;

a prefix of a word within the candidate alias within the candidate aliases;

a postfix of a word within a candidate alias within the candidate aliases; and

a semantic type of a word within a candidate alias within the candidate aliases;

detecting a set of detected entities within the candidate aliases by applying a named entity recognition engine to the candidate aliases, the each detected entity within the set of detected entities comprising a respective feature vector set representing features extracted from the candidate aliases, the applying the named entity recognition engine comprising:

creating a learned model based upon feature vector sets from a first plurality of detected entities; and

applying the learned model to a second plurality of detected entities, the second plurality of detected entities being different than the first plurality;

ranking each of the detected entities according to a respective quality criteria comprising any one of or a combination of any of:

the homogeneity of the detected entity with the input entity;

the occurrence frequency of the tags which contain the detected entity; and

the authoritativeness of the web objects from which the entity is obtained;

collecting detected entities having a rank higher than a predetermined threshold to obtain a collection of aliases of the input entities; and

storing the collection of aliases of the input entities into a dataset.

3. A computer program product comprising a machine readable medium,

the machine readable medium being tangibly encoded with machine executable instructions to perform a method for collecting entity aliases, the method comprising the steps of:

obtaining at least one web object in a result produced by a web search engine in response to a web search query based upon an input entity;

obtaining, from a social tag website in response to obtaining the at least one web object, a respective set of social tags that is associated, on the social tag web site, with the at least one web objects, the respective set of social tags comprising at least one alias for at least one web object within the at least one web objects;

defining, based upon a relationship on the social tag website between social tags within the respective set of social tags and the web objects, candidate aliases for the input entity, the candidate aliases comprising values of social tags within the respective set of social tags;

extracting features from the candidate aliases, the extracting being based upon at least one of:

a word within a candidate alias within the candidate aliases;

at least one of a word preceding and following the candidate alias within the candidate aliases;

capitalization of a beginning letter of a word within the candidate aliases;

a prefix of a word within the candidate alias within the candidate aliases;

a postfix of a word within a candidate alias within the candidate aliases; and

a semantic type of a word within a candidate alias within the candidate aliases;

detecting, using a processor, a set of detected entities within the candidate aliases by applying a named entity recognition engine to the candidate aliases, the each detected entity within the set of detected entities comprising a respective feature vector set representing features extracted from the candidate aliases, the applying the named entity recognition engine comprising:

creating a learned model based upon feature vector sets from a first plurality of detected entities; and

applying the learned model to a second plurality of detected entities, the second plurality of detected entities being different than the first plurality;

ranking each of the detected entities according to a respective quality criteria comprising any one of or a combination of any of:

the homogeneity of the detected entity with the input entity;

the occurrence frequency of the tags which contain the detected entity; and

the authoritativeness of the web objects from which the entity is obtained;

collecting detected entities having a rank higher than a predetermined threshold to obtain a collection of aliases of the input entities; and

storing the collection of aliases of the input entities into a dataset.