# US 12,223,283 B2 · Conversational aspect sentiment analysis for dialogue understanding > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,223,283 B2 - Document ID: US12223283B2 - Status: Granted - Kind: B2 - Application number: 17/503,584 - Filed: 2021-10-18 - Published: 2023-04-20 - Granted: 2025-02-11 - Earliest claimed: 2021-10-18 - Assignees: Tencent America LLC - Applicants: Tencent America LLC - Inventors: Linfeng Song - Web page: https://patentgrep.com/patent/US12223283B2 ## CPC - G06F 40/35: Discourse or dialogue representation - G06F 40/117: Tagging; Marking up (details of markup languages G06F40/143); Designating a block; Setting of attributes (style sheets, e.g. eXtensible Stylesheet Language Transformation [XSLT], G06F40/154) - G06F 40/56: Natural language generation - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 5/041: {Abduction} - G06F 40/284: Lexical analysis, e.g. tokenisation or collocates - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract There is included a method and apparatus for sentiment analysis for multi-turn conversations comprising computer code for obtaining input dialogues; extracting sentiment expressions based on sentence embeddings corresponding to the input dialogues; generating polarity values based on the sentence embeddings corresponding to the input dialogues; and determining a target mention associated with at least one of the sentiment expressions based on the sentiment expressions and the sentence embeddings, wherein the determining of the target mention includes generating rich contextual representations based on the sentence embeddings and the sentiment expressions; and determining the target mention based on calculated boundaries, wherein the calculated boundaries are generated using the rich contextual representations. ## Claims ### Claim 1 A method for sentiment analysis for multi-turn conversations, the method comprising: obtaining an input dialogue, wherein the input dialogue has more than one sentence; extracting at least one sentiment expression based on sentence embeddings corresponding to the input dialogue; generating polarity values corresponding to the at least one sentiment expression based on the sentence embeddings corresponding to the input dialogue; and subsequent to determining the polarity values, determining a target mention associated with the input dialogue based on the at least one sentiment expression and the sentence embeddings, wherein the determining of the target mention comprises: generating rich contextual representations based on a concatenation of turn-wise distances on the sentence embeddings, speaker information on the sentence embeddings, and the at least one sentiment expression, wherein the speaker information indicates whether the input dialogue are from a same speaker; and determining the target mention based on calculated boundaries, wherein the calculated boundaries are generated using the rich contextual representations. ### Claim 2 (depends on 1) The method of claim 1, wherein the determining of the target mention based on calculated boundaries comprises: generating a distribution based on the rich contextual representations and the at least one sentiment expression using one or more attention layers; and determining the target mention based on boundaries of the distribution. ### Claim 3 (depends on 2) The method of claim 2, wherein the generating of the distribution comprises determining a product of distributions of each of the one or more attention layers. ### Claim 4 (depends on 2) The method of claim 2, wherein the determining of the target mention comprises selecting boundaries of the distribution based on a highest score from a plurality of scores, wherein the plurality of scores is generated by determining a product of distributions of each of the one or more attention layers. ### Claim 5 (depends on 1) The method of claim 1, wherein the sentence embeddings are generated based on the input dialogue. ### Claim 6 (depends on 1) The method of claim 1, wherein the extracting the at least one sentiment expression from the sentence embeddings comprises: generating a plurality of tags using a pre-trained machine learned model; and inferring the at least one sentiment expression based on the plurality of tags. ### Claim 7 (depends on 1) The method of claim 1, wherein the generating the polarity values from the sentence embeddings comprises: generating a plurality of tags using a pre-trained machine learned model; and inferring the polarity values based on the plurality of tags. ### Claim 8 (depends on 1) The method of claim 1, wherein the polarity values are one of positive, negative, or neutral. ### Claim 9 An apparatus for sentiment analysis for multi-turn conversations, the apparatus comprising: at least one memory configured to store computer program code; at least one processor configured to access the computer program code and operate as instructed by the computer program code, the computer program code including: first obtaining code configured to cause the at least one processor to obtain input dialogue; first extracting code configured to cause the at least one processor to extract at least one sentiment expression based on sentence embeddings corresponding to the input dialogue; first generating code configured to cause the at least one processor to generate polarity values based on the sentence embeddings corresponding to the input dialogue; and first determining code configured to cause the at least one processor to, subsequent to determining the polarity values, determine a target mention associated with the input dialogue based on the at least one sentiment expression and the sentence embeddings, wherein the first determining code comprises: second generating code configured to cause the at least one processor to generate rich contextual representations based on a concatenation of turn-wise distances on the sentence embeddings, the speaker information on the sentence embeddings, and the at least one sentiment expression, wherein the speaker information indicates whether the input dialogue are from a same speaker; and second determining code configured to cause the at least one processor to determine the target mention based on calculated boundaries, wherein the calculated boundaries are generated using the rich contextual representations. ### Claim 10 (depends on 9) The apparatus of claim 9, wherein the second determining code comprises: fifth generating code configured to cause the at least one processor to generate a distribution based on the rich contextual representations and the at least one sentiment expression using one or more attention layers; and third determining code configured to cause the at least one processor to determine the target mention based on boundaries of the distribution. ### Claim 11 (depends on 10) The apparatus of claim 10, wherein the generating the distribution comprises determining a product of distributions of each of the one or more attention layers. ### Claim 12 (depends on 10) The apparatus of claim 10, wherein the determining the target mention based on the boundaries of the distribution comprises selecting boundaries of the distribution based on a highest score from a plurality of scores, wherein the plurality of scores is generated by determining a product of distributions of each of the one or more attention layers. ### Claim 13 (depends on 9) The apparatus of claim 9, wherein the first extracting code comprises: sixth generating code configured to cause the at least one processor to generate a plurality of tags using a pre-trained machine learned model; and first inferring code configured to cause the at least one processor to infer the at least one sentiment expression based on the plurality of tags. ### Claim 14 (depends on 9) The apparatus of claim 9, wherein the first generating code comprises: sixth generating code configured to cause the at least one processor to generate a plurality of tags using a pre-trained machine learned model; and first inferring code configured to cause the at least one processor to infer the at least one sentiment expression based on the plurality of tags. ### Claim 15 (depends on 9) The apparatus of claim 9, wherein the polarity values are one of positive, negative, or neutral. ### Claim 16 A non-transitory computer readable medium storing a program causing a computer to execute a process, the process comprising: obtaining input dialogue; extracting at least one sentiment expression based on sentence embeddings corresponding to the input dialogue; generating polarity values based on the sentence embeddings corresponding to the input dialogue; and subsequent to determining the polarity values, determining a target mention associated with the input dialogue based on the at least one sentiment expression and the sentence embeddings, wherein the determining of the target mention comprises: generating rich contextual representations based on a concatenation of turn-wise distances on the sentence embeddings, the speaker information on the sentence embeddings, and the at least one sentiment expression, wherein the speaker information indicates whether the input dialogue are from a same speaker; and determining the target mention based on calculated boundaries, wherein the calculated boundaries are generated using the rich contextual representations. ### Claim 17 (depends on 16) The non-transitory computer readable medium of claim 16, wherein the determining of the target mention based on calculated boundaries comprises: generating a distribution based on the rich contextual representations and the at least one sentiment expression using one or more attention layers; and determining the target mention based on boundaries of the distribution.