# US 12,182,515 B2 · Utterance intent detection > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,182,515 B2 - Document ID: US12182515B2 - Status: Granted - Kind: B2 - Application number: 17/389,701 - Filed: 2021-07-30 - Published: 2023-02-02 - Granted: 2024-12-31 - Earliest claimed: 2021-07-30 - Assignees: Intuit Inc. - Applicants: Intuit Inc. - Inventors: Yan Qu; Lin Tao; Yingxin Wang - Web page: https://patentgrep.com/patent/US12182515B2 ## CPC - G06F 40/30: Semantic analysis - G06F 16/3344: {using natural language analysis} - G06F 16/3347: {using vector based model} - G06F 16/338: Presentation of query results - G06F 40/44: Statistical methods, e.g. probability models - G06N 3/045: Combinations of networks - G06N 3/0475: Generative networks - G06N 3/08: Learning methods - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Certain aspects of the present disclosure provide techniques for detecting sentences that are utterances by an agent indicating an intent to poach a customer. According to certain embodiments, a language model is trained using query sentences that are confirmed to be sentences used in poaching a customer, to identify semantically similar sentences in a corpus. These semantically similar sentences are then used as base sentences for comparison to sentences in a transcript. Sentences of the transcript that are found to be semantically similar to one or more base sentences are provided to a user device for review and confirmation that the similar sentence was generated by an agent in an attempt to poach a customer. ## Claims ### Claim 1 A method, comprising: receiving a plurality of corpus sentences, wherein each corpus sentence of the plurality of corpus sentences comprises a sentence from a respective transcript of previous interactions involving a respective agent; receiving a plurality of query sentences, wherein each query sentence of the plurality of query sentences is associated with an intent to poach a customer; determining a query sentence embedding for each query sentence of the plurality of query sentences; determining a corpus sentence embedding for each corpus sentence of the plurality of corpus sentences, based on one or more query sentence embeddings of the plurality of query sentences; determining a semantic similarity score for at least one corpus sentence embedding relative to at least one query sentence embedding of the plurality of query sentences; identifying one corpus sentence as a base sentence, based on the semantic similarity score associated with the corpus sentence embedding of the corpus sentence being at or above a threshold; receiving a transcript sentence; providing the transcript sentence to a machine learning model, wherein the machine learning model has been trained through a supervised learning process using training data that includes the base sentence to determine a transcript similarity score for the transcript sentence based on the base sentence; receiving a transcript similarity score for the transcript sentence from the machine learning model; detecting the intent in the transcript sentence based on the transcript similarity score exceeding a threshold; and providing an output to a user device based on the detected intent. ### Claim 2 (depends on 1) The method of claim 1, wherein determining the corpus sentence embedding for each corpus sentence of the plurality of corpus sentences comprises clustering each corpus sentence embedding based on one or more of the query sentence embeddings. ### Claim 3 (depends on 2) The method of claim 2, wherein the clustering is carried out by one of a bidirectional encoder representations from transformers (BERT) or a sentence BERT (SBERT). ### Claim 4 (depends on 3) The method of claim 3, wherein determining the semantic similarity score comprises determining a cosine similarity between the at least one corpus sentence embedding and the at least one query sentence embedding. ### Claim 5 (depends on 3) The method of claim 3, wherein one of the plurality of corpus sentences is split into two or more corpus sentences. ### Claim 6 (depends on 5) The method of claim 5, wherein the output provided to the user device identifies the transcript similarity score and the transcript sentence for human review. ### Claim 7 (depends on 6) The method of claim 6, wherein the transcript is generated from one of a voice or text interaction between an agent and a customer. ### Claim 8 A system comprising: a memory comprising computer-readable instructions; a processor configured to read the computer-readable instructions, that cause the system to: receive a plurality of corpus sentences, wherein each corpus sentence of the plurality of corpus sentences comprises a sentence from a respective transcript of previous interactions involving a respective agent; receive a plurality of query sentences, wherein each query sentence of the plurality of query sentences is associated with an intent to poach a customer; determine a query sentence embedding for each query sentence of the plurality of query sentences; determine a corpus sentence embedding for each corpus sentence of the plurality of corpus sentences, based on one or more query sentence embeddings of the plurality of query sentences; determine a semantic similarity score for at least one corpus sentence embedding relative to at least one query sentence embedding of the plurality of query sentences; identify one corpus sentence as a base sentence, based on the semantic similarity score associated with the corpus sentence embedding of the corpus sentence being at or above a threshold; receive a transcript sentence; provide the transcript sentence to a machine learning model, wherein the machine learning model has been trained through a supervised learning process using training data that includes the base sentence to determine a transcript similarity score for the transcript sentence, based on the base sentence; receive a transcript similarity score for the transcript sentence from the machine learning model; detect the intent in the transcript sentence based on the transcript similarity score exceeding a threshold; and provide an indicator of the transcript similarity score to a user device. ### Claim 9 (depends on 8) The system of claim 8, wherein determining the corpus sentence embedding for each corpus sentence of the plurality of corpus sentences comprises clustering each corpus sentence embedding based on one or more of the query sentence embeddings. ### Claim 10 (depends on 9) The system of claim 9, wherein the clustering is carried out by one of a bidirectional encoder representations from transformers (BERT) or a sentence BERT (SBERT). ### Claim 11 (depends on 10) The system of claim 10, wherein determining the semantic similarity score comprises determining a cosine similarity between the at least one corpus sentence embedding and the at least one query sentence embedding. ### Claim 12 (depends on 10) The system of claim 10, wherein one of the plurality of corpus sentences is split into two or more corpus sentences. ### Claim 13 (depends on 12) The system of claim 12, wherein the indicator identifies the transcript similarity score and the transcript sentence for human review. ### Claim 14 (depends on 13) The system of claim 13, wherein the transcript is generated from one of a voice or text interaction between an agent and a customer. ### Claim 15 A method comprising: receiving a base sentence embedding, wherein the base sentence embedding is generated for a base sentence, wherein the base sentence embedding is used to identify the base sentence as semantically similar to a sentence that is associated with an intent to poach a customer; receiving agent content; generating an agent content embedding from the agent content; generating a similarity score between the base sentence embedding and the agent content embedding using a machine learning model that has been trained through a supervised learning process using training data that includes the base sentence; and identifying the intent and providing an output to a user based on the similarity score exceeding a given threshold. ### Claim 16 (depends on 15) The method of claim 15, further comprising: receiving a plurality of corpus sentences, wherein each corpus sentence of the plurality of corpus sentences comprises a sentence from a respective transcript of previous interactions involving a respective agent; receiving a plurality of query sentences, wherein each query sentence of the plurality of query sentences is associated with an intent to poach a customer; determining a corpus sentence embedding for each corpus sentence of the plurality of corpus sentences, based on one or more query sentence embeddings of the plurality of query sentences; detecting the intent in at least one corpus sentence of the plurality of corpus sentences based on providing the at least one corpus sentence to a machine learning model, wherein the machine learning model has been trained through a supervised learning process using training data that includes the base sentence to determine a semantic similarity score for the corpus sentence embedding of the at least one corpus sentence relative to at least one query sentence embedding of the plurality of query sentences; and identifying the at least one corpus sentence as an additional base sentence, based on the semantic similarity score associated with the corpus sentence embedding of the at least one corpus sentence being at or above a threshold. ### Claim 17 (depends on 16) The method of claim 16, wherein determining the corpus sentence embedding for each corpus sentence of the plurality of corpus sentences comprises clustering each corpus sentence embedding based on one or more of the query sentence embeddings. ### Claim 18 (depends on 17) The method of claim 17, wherein the clustering is carried out by one of a bidirectional encoder representations from transformers (BERT) or a sentence BERT (SBERT). ### Claim 19 (depends on 15) The method of claim 15 further comprising: receiving a filter sentence; receiving customer content comprising a customer sentence; and matching the filter sentence to the customer sentence. ### Claim 20 (depends on 19) The method of claim 19, wherein generating the similarity score further comprises generating the similarity score based on the matching of the filter sentence to the customer sentence.