Systems and methods of natural language processing to rank users of real time communications connections
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10412225-B2
Publication Date
2019-09-10
Expiration Date
Abstract
A computing apparatus configured to perform natural language processing, e.g., by comparing the words in a communication transcript of a user to a list of keywords, to generate an input vector representing a pattern of text in the communication transcript. A predictive model is generated from correlating input vectors to user ranking scores, e.g., for retention. The input vector determined from a communication transcript is applied to the computation model to compute a predicted retention score of the user. The retention score can be used, for example, to select a personalized recommendation for a communication connection to an adviser and/or a targeted offer.
Core Innovation
The invention addresses a real-time communication marketplace in which advisers are connected to customers via a connection server, with communication connections established between telephonic devices using communication references. It stores transcripts of communications over the communication connections in a computing apparatus, so that communication records can be used for downstream ranking and management.
A core part of the system performs natural language processing on the stored communication transcripts to generate input vectors. The input vectors are derived from communication transcript content, and they are correlated with ranking scores of the second users generated from the communication records to establish a predictive model of ranking score.
The documented implementations further include deriving transcript-based representations using voice recognition signals and natural language processing features such as text patterns and keyword list processing reflected by term frequencies. The resulting predictive model supports ranking behavior by using statistical correlation analysis and machine learning to align predicted ranking scores with ranking scores from communication records.
Claims Coverage
The provided set includes three independent claims that share the same core workflow: connecting telephonic devices via a connection server, storing communication transcripts, applying natural language processing to generate input vectors, and generating ranking scores using a predictive model where the input vectors are correlated with the ranking scores. The dependent claim members refine key inventive components such as transcript generation, keyword-list-based NLP, term-frequency input vectors, and statistical correlation and machine-learning training.
Connection server bridging with stored communication transcripts
In response to requests received from second users identifying selected ones of first users, a connection server establishes communication connections between respective telephonic devices of the second users and respective telephonic devices of the selected ones of the first users using communication references, and a computing apparatus stores transcripts of communications over the communication connections.
Natural language processing on transcripts to generate input vectors
The computing apparatus performs natural language processing on the stored transcripts to generate input vectors.
Predictive model of ranking score from correlated input vectors and communication records
The computing apparatus generates ranking scores of the second users from the communication records, wherein the input vectors are correlated with the ranking scores to establish a predictive model of ranking score.
Voice-recognition transcript generation
The method further includes generating a transcript of the communication by performing voice recognition on the communication transmitted through a connection server that bridges a first connection and a second connection.
Keyword-list-based natural language processing
The computing apparatus performs natural language processing based on a list of stored keywords.
Term-frequency input vectors for keyword list features
The generated input vectors include term frequencies of the listed keywords in the transcripts.
Statistical correlation and machine-learning-based predictive modeling
The method uses machine learning during training to build a predictive ranking-score model by correlating input vectors with ranking scores so the model predicts scores matching the ranking scores from the input vectors.
Downstream recommendation based on ranking score
The method further includes using a computing apparatus to identify a recommended connection to one of the first users based at least in part on the user’s ranking score.
Across the independent claims, the inventive coverage centers on using transcript-derived natural language processing input vectors correlated with ranking scores to build a predictive model that generates ranking scores for second users based on communication records from connection-server-established telephonic communications. The listed dependent claim members further sharpen this with voice recognition, keyword lists, term-frequency representations, statistical correlation, machine-learning training, and recommending connections based on ranking scores.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?