Methods and systems for adaptive collaborative matching
An adaptive collaborative platform applies various machine learning techniques to correlate potential purchasers with high-value articles of property that may be of interest. Attributes, characteristics, preferences, and the like of a potential purchaser are scored against attributes and features of articles. The platform learns from interaction by the agents and the potential purchasers and adapts to become more attuned to the desires and lifestyle of purchasers and to gain more and more pertinent information from the listing agents regarding high-value articles, so as to ultimately to arrive at a better match between a high value article for sale and a likely purchaser.
1 . A machine implemented method for matching correlated entities, the method comprising:
receiving, via a server, structured empirical data and unstructured data regarding each of a plurality of entities from a plurality of sources;
converting, using natural language processing and semantic analysis, the unstructured data into structured empirical data conforming to a predefined schema;
storing the structured empirical data in a data store organized into data fields grouped by factors, each factor assigned a weight;
in response to identifying, by server, one or more first missing data fields in the structured empirical data, automatically accessing a third party website and retrieving data corresponding to the one or more first missing data fields;
in response to retrieving the data corresponding to the one or more first missing data fields, identifying one or more second missing data fields in the structured empirical data and automatically accessing a second third party website and retrieving data corresponding to the one or more second missing data fields;
deriving, by a tag derivation engine of the server, a plurality of tags, each tag comprising a discrete grouping of factors with factor weights, and assigning to each entity one or more tags with associated tag confidence scores;
deriving, by a lifestyle engine of the server, one or more lifestyle scores for each entity based on weighted combinations of the tags and the tag confidence scores;
correlating, by a matching engine of the server, the plurality of entities based on the one or more lifestyle scores to generate a first matching model;
displaying, on an agent device, first ranked correlations of the plurality of entities and receiving user feedback from an agent via the agent device;
refining, by the matching engine, the first matching model to form a second matching model based on the user feedback received through a first graphical user interface displayed on the agent device, wherein the second matching model adaptively modifies the factor weights and the one or more tags assigned to each entity; and
in response to receiving a query, via a client device, displaying, on a second graphical user interface displayed on the client device, second ranked correlations of the plurality of entities exceeding a predefined accuracy threshold, wherein the second ranked correlations are determined by the second matching model and are based at least partially on the user feedback.
2 . The method of claim 1 , wherein:
the matching engine refines the first matching model into the second matching model by modifying the factor weights or the one or more tags assigned to each entity responsive to user feedback scores; and
adopts the second matching model in response to determining that a correlation accuracy of the second matching model exceeds a correlation accuracy of the first matching model.
3 . The method of claim 1 , further comprising:
ranking the plurality of correlated entities by a lifestyle score similarity; and
displaying only correlated entities of the plurality of correlated entities that exceed the predefined accuracy threshold on the second graphical user interface.
4 . The method of claim 1 , further comprising:
transmitting, by a normalization engine of the server, an inquiry to the agent requesting data corresponding to unresolved missing data fields; and
updating the structured empirical data in the data store with the data corresponding to the unresolved missing data fields.
5 . The method of claim 1 , wherein a normalization engine of the server:
applies the natural language processing to extract quantitative attributes from the unstructured textual data; and
maps the extracted quantitative attributes into the predefined schema.
6 . An adaptive collaborative matching system, comprising:
a processor; and
a non-transitory memory storing instructions that, when executed by the processor, configure the system to:
receive, via a server, structured empirical data and unstructured data for a plurality of entities from a plurality of sources;
convert, using natural language processing and semantic analysis, the unstructured data into structured empirical data conforming to a predefined schema, and store the structured empirical data in a data store;
group the structured empirical data into a plurality of data fields organized by factors, each factor assigned a factor weight;
in response to identifying, by server, one or more first missing data fields in the structured empirical data, automatically accessing a third party website and retrieving data corresponding to the one or more first missing data fields;
in response to retrieving the data corresponding to the one or more first missing data fields, identifying one or more second missing data fields in the structured empirical data and automatically accessing a second third party website and retrieving data corresponding to the one or more second missing data fields;
derive, by a tag derivation engine of the server, a plurality of tags, each tag comprising a discrete grouping of factors with factor weights, and assign to each entity one or more tags with associated tag confidence scores;
derive, by a lifestyle engine of the server, one or more lifestyle scores for each entity based on weighted combinations of the tags and the tag confidence scores;
correlate, by a matching engine of the server, the plurality of entities based on the one or more lifestyle scores to generate a first matching model;
displaying, on an agent device, a first ranked list of correlated entities of the plurality of entities and receiving user feedback from an agent via the agent device;
refine, by the matching engine, the first matching model into a second matching model by modifying the factor weights and the one or more tags assigned to each entity based on the user feedback received via a first graphical user interface displayed on the agent device; and
in response to receiving a query, via a client device, display, on a second graphical user interface displayed on the client device, a second ranked list of correlated entities exceeding a predefined correlation threshold, wherein the second ranked list of correlated entities are determined by the second matching model and are based at least partially on the user feedback.
7 . The system of claim 6 , wherein a normalization engine of the server is further configured to transmit a query to the agent requesting additional data to resolve missing fields.
8 . The system of claim 6 , wherein the tag derivation engine applies the natural language processing and the semantic analysis to extract attributes from the unstructured data prior to deriving the tags.
9 . The system of claim 6 , wherein the lifestyle engine generates a plurality of lifestyle scores for each entity, each lifestyle score corresponding to one of a predetermined set of lifestyle categories.
10 . The system of claim 6 , wherein the matching engine is further configured to rank the plurality of entities based on a lifestyle score similarity and discard correlated entities of the plurality of correlated entities not exceeding a minimum threshold.
11 . The system of claim 6 , wherein the processor is further configured to generate, in response to the user feedback, a modified entity profile accessible only to the agent, the modified entity profile including enriched structured data.
12 . The system of claim 6 , wherein the processor is further configured to iteratively update the second matching model into subsequent matching models in response to determining that the user feedback indicates improved correlation accuracy.
13 . The system of claim 6 , wherein the first graphical user interface is configured to:
receive agent input modifying the factor weights or the one or more tags assigned to each entity; and
update the one or more lifestyle scores and the first matching model in response to the agent input.