Contextualized attributes for vectorized data
A method that includes at least one of receiving interaction content from a communication session between a source device and a service provider device of a service provider, identifying a plurality of contextual attributes of the communication session based on execution of at least one large language models (LLMs) on the interaction content, converting the interaction content and the plurality of contextual attributes of the communication session into vectorized data based on execution of an additional LLM, labelling the vectorized data with identifiers of the plurality of contextual attributes, and storing the vectorized data within a vector database.
1 . An apparatus comprising:
a memory storing instructions; and
a processor coupled to the memory and configured to execute the instructions, wherein when executed, the instructions configure the processor to:
receive conversation content from a communication session between a source device and a provider device,
identify a plurality of contextual attributes of the communication session based on execution of at least one large language models (LLMs) on the conversation content,
annotate the conversation content to include identifiers of the plurality of contextual attributes therein,
convert the annotated conversation content into a vector based on execution of an additional LLM,
add a label to the vector with identifiers of the plurality of contextual attributes used to annotate the conversation content, and
store the labelled vector within a vector database based on a contextual attribute in the label.
2 . The apparatus of claim 1 , wherein the processor is configured to store the vector within a storage area of the vector database dedicated to the source device which includes at least other vectors from communication session with the source device.
3 . The apparatus of claim 1 , wherein the processor is configured to annotate the conversation content with identifiers of the plurality of contextual attributes at temporal locations within the conversation content at which the plurality of contextual attributes are detected.
4 . The apparatus of claim 1 , wherein the processor is configured to identify an item of interest and a mood with respect to the item of interest from the conversation content, and label the vector with an identifier of the item of interest and an additional identifier of the mood.
5 . The apparatus of claim 1 , wherein the processor is configured to identify a date and a time at which the communication session occurred, and label the vector with an identifier of the date and the time at which the communication session occurred.
6 . The apparatus of claim 1 , wherein the processor is further configured to receive a search query from a software application, identify at least one vector within the vector database that correspond to the search query based on labels of the at least one vector, and transmit the at least one vector to the software application.
7 . The apparatus of claim 1 , wherein the processor is further configured to identify a different vector stored within the vector database more than a predetermined period of time based on a comparison of a respective timestamp of the different vector to a current time, and in response, remove the different vector from the vector database.
8 . The apparatus of claim 1 , wherein the processor is configured to identifying the plurality of contextual attributes based on execution of a LLM comprising a plurality of attention heads corresponding to the plurality of contextual attributes, respectively.
9 . A method comprising:
receiving, by hardware processor, conversation content from a communication session between a source device and a provider device;
identifying a plurality of contextual attributes of the communication session based on execution of at least one large language models (LLMs) by the hardware processor on the conversation content;
annotating, by the hardware processor, the conversation content to include identifiers of the plurality of contextual attributes therein;
converting, by the hardware processor, the annotated conversation content into a vector based on execution of an additional LLM;
adding, by the hardware processor, a label to the vector with identifiers of the plurality of contextual attributes used to annotate the conversation content; and
storing, by the hardware processor, the labelled vector within a vector database based on a contextual attribute in the label.
10 . The method of claim 9 , wherein the storing comprises storing the vector within a storage area of the vector database dedicated to the source device which includes at least other vectors from communication session with the source device.
11 . The method of claim 9 , wherein the annotating comprises annotating the conversation content with identifiers of the plurality of contextual attributes at temporal locations within the conversation content at which the plurality of contextual attributes are detected.
12 . The method of claim 9 , wherein the identifying comprises identifying an item of interest and a mood with respect to the item of interest from the conversation content, and the adding comprises labelling the vector with an identifier of the item of interest and an additional identifier of the mood.
13 . The method of claim 9 , wherein the identifying comprises identifying a date and a time at which the communication session occurred, and the adding comprises labelling vector with an identifier of the date and the time at which the communication session occurred.
14 . The method of claim 9 , wherein the method further comprises receiving a search query from a software application, identifying at least one vector within the vector database that correspond to the search query based on labels of the at least one vector, and transmitting the at least one vector to the software application.
15 . The method of claim 9 , wherein the adding comprises adding metadata to the vector which includes the identifiers of the plurality of contextual attributes, prior to storing the vector in the vector database.
16 . The method of claim 9 , wherein the method further comprises identifying a different vector stored within the vector database more than a predetermined period of time based on a comparison of a respective timestamp of the different vector to a current time, and in response, removing the different vector from the vector database.
17 . A non-transitory computer-readable storage medium comprising instructions stored therein which when executed by a processor cause the processor to perform:
receiving conversation content from a communication session between a source device and a provider device;
identifying a plurality of contextual attributes of the communication session based on execution of at least one large language models (LLMs) on the conversation content;
annotating the conversation content to include identifiers of the plurality of contextual attributes therein;
converting the annotated conversation content into a vector based on execution of an additional LLM;
adding a label to the vector with identifiers of the plurality of contextual attributes used to annotate the conversation content; and
storing the labelled vector within a vector database based on a contextual attribute in the label.
18 . The non-transitory computer-readable storage medium of claim 17 , wherein the storing comprises storing the vector within a storage area of the vector database dedicated to the source device.
19 . The non-transitory computer-readable storage medium of claim 17 , wherein the annotating comprises annotating the conversation content with identifiers of the plurality of contextual attributes at temporal locations within the conversation content at which the plurality of contextual attributes are detected.
20 . The non-transitory computer-readable storage medium of claim 17 , wherein the identifying comprises identifying an item of interest and a mood with respect to the item of interest from the conversation content, and the adding comprises labelling the vector with an identifier of the item of interest and an additional identifier of the mood.