# US 10,769,378 B2 · Extending system entities for conversational system > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,769,378 B2 - Document ID: US10769378B2 - Status: Granted - Kind: B2 - Application number: 16/190,581 - Filed: 2018-11-14 - Published: 2020-05-14 - Granted: 2020-09-08 - Earliest claimed: 2018-11-14 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Li Jun Mei; Qi Cheng Li; Rangachari Anand; Ajay Gupta - Web page: https://patentgrep.com/patent/US10769378B2 ## CPC - G06F 40/295: Named entity recognition - G06F 40/35: Discourse or dialogue representation - G06F 16/2465: {Query processing support for facilitating data mining operations in structured databases} - G06F 40/216: using statistical methods - G06F 40/284: Lexical analysis, e.g. tokenisation or collocates - G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking - G06F 40/30: Semantic analysis - G06N 5/025: Extracting rules from data - G06N 5/04: Inference or reasoning models - G06N 20/00: Machine learning - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A computer-implemented method is provided for processing a set D of conversation logs. The method includes learning, by a processor device, a set R of entity relation mining rules from a set K of known entity utterances uttered by known entities. The method further includes automatically recommending, by the processor device, extended utterances of the known entities from unrecognized ones of the known entities based on the set R. ## Claims ### Claim 1 A computer-implemented method for processing a set D of conversation logs, comprising: learning, by a processor device, a set R of entity relation mining rules from a set K of known entity utterances uttered by known entities; and automatically recommending, by the processor device, extended utterances of the known entities from unrecognized ones of the known entities based on the set R; wherein said learning step comprises: replacing, in the set D of conversation logs, any of the known entity utterances from the set K that appear in the set D, by a respective representative utterance having a same meaning to form a set D1 of unified conversation logs; finding a set CS of chat log segments in the set D1 which include at least two of the same utterances in the set D1; generating the set R of entity relation mining rules based on the set CS; randomly selecting a subset K1 from the set K; for each of the respective representative utterance in the set D1 that also occurs in the subset K1, replacing the respective representative utterance with a corresponding original utterance from the set D to form a list D2 of conversation logs; removing the subset K1 from the set K; and for each individual rule r in the set R, applying the individual rule r on the list D2, checking a success rate of identifying utterance relations in the subset K1, and removing the individual rule r from the set R if the success rate is below a rule success rate threshold. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, wherein said first replacing step comprises maintaining mapping information between each of the known entity utterances from the set K and the respective representative utterance therefor, and wherein said second replacing step replaces the respective representative utterance with the corresponding original utterance from the set D using the mapping information to form the list D2. ### Claim 3 (depends on 1) The computer-implemented method of claim 1, wherein the respective representative utterance having the same meaning is comprised in a set S of respective representative utterances having same meanings to expected words in the set D. ### Claim 4 (depends on 1) The computer-implemented method of claim 1, wherein said recommending step comprises: generating a set of word embeddings from the set D; generating a set of phrase embeddings from the set of word embeddings; identifying words and phrases having similar embedding vectors relative to a similarity threshold; ranking candidate words and phrases based on character-level similarity; and recommending at least some of the candidate words and phrases as the extended utterances. ### Claim 5 (depends on 4) The computer-implemented method of claim 4, further comprising: determining whether any of the candidate words and phrases match any of the rules in the set R; recommending matching ones of the candidate words and phrases to a subject matter expert for verification; and discarding non-matching ones of the candidate words and phrases. ### Claim 6 (depends on 4) The computer-implemented method of claim 4, wherein the word embedding and the phrase embeddings are used to obtain distributed representation vectors for both the known entities and unknown entities. ### Claim 7 (depends on 1) The computer-implemented method of claim 1, wherein the set R is learned relative to a particular task, and wherein the known entity utterances specifically relate to the particular task. ### Claim 8 (depends on 1) The computer-implemented method of claim 1, further comprising controlling a processor-based machine to perform an action responsive to the extended utterance. ### Claim 9 (depends on 8) The computer-implemented method of claim 8, wherein the processor-based machine is an Advanced Driver-Assistance System, and the extended utterance relates to controlling a trajectory of a vehicle. ### Claim 10 (depends on 1) The computer-implemented method of claim 1, wherein the method is provided as a cloud-based service. ### Claim 11 A computer program product for processing a set D of conversation logs, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising: learning, by a processor device, a set R of entity relation mining rules from a set K of known entity utterances uttered by known entities; and automatically recommending, by the processor device, extended utterances of the known entities from unrecognized ones of the known entities based on the set R; wherein said learning step comprises: replacing, in the set D of conversation logs, any of the known entity utterances from the set K that appear in the set D, by a respective representative utterance having a same meaning to form a set D1 of unified conversation logs; finding a set CS of chat log segments in the set D1 which include at least two of the same utterances in the set D1; generating the set R of entity relation mining rules based on the set CS; randomly selecting a subset K1 from the set K; for each of the respective representative utterance in the set D1 that also occurs in the subset K1, replacing the respective representative utterance with a corresponding original utterance from the set D to form a list D2 of conversation logs; removing the subset K1 from the set K; and for each individual rule r in the set R, applying the individual rule r on the list D2, checking a success rate of identifying utterance relations in the subset K1, and removing the individual rule r from the set R if the success rate is below a rule success rate threshold. ### Claim 12 (depends on 11) The computer program product of claim 11, wherein said first replacing step comprises maintaining mapping information between each of the known entity utterances from the set K and the respective representative utterance therefor, and wherein said second replacing step replaces the respective representative utterance with the corresponding original utterance from the set D using the mapping information to form the list D2. ### Claim 13 (depends on 11) The computer program product of claim 11, wherein the respective representative utterance having the same meaning is comprised in a set S of respective representative utterances having same meanings to expected words in the set D. ### Claim 14 (depends on 11) The computer program product of claim 11, wherein said recommending step comprises: generating a set of word embeddings from the set D; generating a set of phrase embeddings from the set of word embeddings; identifying words and phrases having similar embedding vectors relative to a similarity threshold; ranking candidate words and phrases based on character-level similarity; and recommending at least some of the candidate words and phrases as the extended utterances. ### Claim 15 (depends on 14) The computer program product of claim 14, wherein the method further comprises: determining whether any of the candidate words and phrases match any of the rules in the set R; recommending matching ones of the candidate words and phrases to a subject matter expert for verification; and discarding non-matching ones of the candidate words and phrases. ### Claim 16 (depends on 15) The computer program product of claim 15, wherein the word embedding and the phrase embeddings are used to obtain distributed representation vectors for both the known entities and unknown entities. ### Claim 17 (depends on 11) The computer program product of claim 11, wherein the set R is learned relative to a particular task, and wherein the known entity utterances specifically relate to the particular task. ### Claim 18 (depends on 11) The computer program product of claim 11, wherein the method further comprises controlling a processor-based machine to perform an action responsive to the extended utterance. ### Claim 19 (depends on 18) The computer program product of claim 18, wherein the processor-based machine is an Advanced Driver-Assistance System, and the extended utterance relates to controlling a trajectory of a vehicle. ### Claim 20 A computer processing system for processing a set D of conversation logs, comprising: a memory for storing program code; and a processor device for running the program code to learn a set R of entity relation mining rules from a set K of known entity utterances uttered by known entities; and automatically recommend extended utterances of the known entities from unrecognized ones of the known entities based on the set R; wherein the set R is learned by: replacing, in the set D of conversation logs, any of the known entity utterances from the set K that appear in the set D, by a respective representative utterance having a same meaning to form a set D1 of unified conversation logs; finding a set CS of chat log segments in the set D1 which include at least two of the same utterances in the set D1; generating the set R of entity relation mining rules based on the set CS; randomly selecting a subset K1 from the set K; for each of the respective representative utterance in the set D1 that also occurs in the subset K1, replacing the respective representative utterance with a corresponding original utterance from the set D to form a list D2 of conversation logs; removing the subset K1 from the set K; and for each individual rule r in the set R, applying the individual rule r on the list D2, checking a success rate of identifying utterance relations in the subset K1, and removing the individual rule r from the set R if the success rate is below a rule success rate threshold.