Tagging relations with N-best

A non-transitory computer-readable storage medium in which computer program instructions are stored, which instructions, when executed by a computer, cause the computer to perform a method of automated development of an ontology by detecting and tagging relations in a set of communication data, wherein the ontology is a structural representation of language elements and relationships between those language elements within a domain stored in memory of the computer system, the method including the steps comprising:

• identifying terms in the set of communication data, wherein the set of communication data comprises audio data from one or more conversations;
• identifying possible relations of the identified terms, wherein each of the possible relations is a binary directed relationship between a pair of the identified terms;
• scoring the possible relations according to a set of predefined merits, including assigning a relatively higher score to one or more of the possible relations whose terms have higher average transcription scores, wherein each of the transcription scores is determined based on selecting a best path out of a word lattice or confusion network of a predetermined number of best potential transcriptions output from a large vocabulary continuous speech recognition (LVCSR) transcription of the audio data;
• ranking the possible relations into a ranked list of the possible relations in descending order according to their score;
• tagging the possible relations in the set of communication data, wherein the possible relations are tagged by identifying the possible relations in the set of communication data in order corresponding with the ranked list of the possible relations, wherein lower ranking possible relations that conflict with higher ranking relations according to a predefined set of conflicts are not tagged;
• and
• outputting an ontological structure based on the tagged relations to a storage medium for application to incoming communication data to analyze and extract business data from the incoming communication data.

An apparatus for automated development of an ontology, wherein the ontology is a structural representation of language elements and relationships between those language elements within a domain stored in memory, comprising:

• a communication interface configured to receive communication data, the communication data comprising audio data from one or more conversations;
• a hardware processor configured to perform the steps comprising:
• identifying terms in the received communication data,
• identifying possible relations of the identified terms, wherein each of the possible relations is a binary directed relationship between a pair of the identified terms;
• scoring the possible relations according to a set of predefined merits, including assigning a relatively higher score to one or more of the possible relations whose terms have higher average transcription scores, wherein each of the transcription scores is determined based on selecting a best path out of a word lattice or confusion network of a predetermined number of best potential transcriptions output from a large vocabulary continuous speech recognition (LVCSR) transcription of the audio data,
• ranking the possible relations into a list of the possible relations in descending order according to their score,
• tagging the possible relations in the received communication data, wherein the relations are tagged by identifying the possible relations in the communication data in order corresponding with the list of the possible relations, wherein lower ranking possible relations that conflict with higher ranking relations according to a predefined set of conflicts are not tagged, and
• outputting an ontological structure based on the tagged relations;
• and
• a user interface configured to output the tagged relations.

A method of automated development of an ontology via a computer system detecting and tagging relations in a set of communication data, wherein the ontology is a structural representation of language elements and relationships between those language elements within a domain stored in memory of the computer system, the method comprising:

• identifying terms in the set of communication data, wherein the set of communication data comprises audio data from one or more conversations;
• identifying possible relations of the identified terms, wherein each of the possible relations is a binary directed relationship between an identified term and an entity or subclass or between an entity and a subclass;
• scoring the possible relations according to a set of predefined merits, including assigning a relatively higher score to one or more of the possible relations whose terms have higher average transcription scores, wherein each of the transcription scores is determined based on selecting a best path out of a word lattice or confusion network of a predetermined number of best potential transcriptions output from a large vocabulary continuous speech recognition (LVCSR) transcription of the audio data;
• ranking the possible relations into a ranked list of the possible relations in descending order according to their score;
• tagging the possible relations in the set of communication data, wherein the possible relations are tagged by identifying the possible relations in the set of communication data in order corresponding with the ranked list of the possible relations, wherein lower ranking possible relations that conflict with higher ranking relations according to a predefined set of conflicts are not tagged;
• and
• outputting an ontological structure based on the tagged relations to a storage medium for application to incoming communication data to analyze and extract business data from the incoming communication data.