US 11,947,604 B2
Ranking of messages in dialogs using fixed point operations
Haggai Roitman, Yokneam Illit (IL); Doron Cohen, Gilon (IL); Yosi Mass, Ramat Gan (IL); and Shai Erera, Gilon (IL)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Mar. 17, 2020, as Appl. No. 16/821,071.
Prior Publication US 2021/0294863 A1, Sep. 23, 2021
Int. Cl. G06F 16/953 (2019.01); G06F 16/901 (2019.01); G06F 40/30 (2020.01); G06Q 30/016 (2023.01)
CPC G06F 16/953 (2019.01) [G06F 16/9024 (2019.01); G06F 40/30 (2020.01); G06Q 30/016 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising a processor, a computer-readable storage media, and program instructions stored on the computer-readable storage media for causing the processor to:
receive a dialog comprising multiple messages generated from multiple computers and exchanged between the multiple computers, the exchanges occurring within a fixed time length corresponding to the dialog;
concatenating the received messages to form a verbose query;
transmit the verbose query to a first search engine;
receive a pseudo-relevance set comprising higher ranked results from the first search engine in response to the transmitting;
execute a first fixed point operation on the pseudo-relevance set to generate weighted terms, wherein the first fixed point operation comprises an iterative process in which a vector including scores associated with the weighted terms is processed until values of the vector remain stable within a range of error, wherein term weights of the weighted terms are temporally biased using a function that decays a weight of each query term of the verbose query based on an index of the last message of the dialog in which a respective query term appears, wherein the index temporally orders the set of concatenated messages, and wherein the term weights are further determined based on frequency of the respective query term within the dialog;
generate a final message ranking by executing a second fixed point operation on a message graph comprising nodes representing messages of the dialog with a heaviness based on the weighted terms and edges connecting the nodes, the edges including weights indicating a similarity between connected messages of the dialog;
generate a weighted query based on the final message ranking, wherein each message in the dialog is treated as a sub-query with a weight in the weighted query, wherein the processor filters out messages from the set of concatenated message of the dialog based on the final message ranking to generate a set of filtered messages and to generate the weighted query based on the set of filtered messages;
transmit the weighted query to a second search engine, wherein the second search engine executes a contextual multi-field search using the weighted query as an input query into the contextual multi-field search; and
receive and present search results generated by the weighted query being the input query in the contextual multi-field search of the second search engine.