Real time recommender modeling system, methods of construction, and methods of use

A non-transitory, computer-readable storage medium having encoded thereon machine instructions to implement a real-time recommender method, wherein a processor executes the machine instruction to:

• receive streaming data from a plurality of streaming data sources, the streaming data comprising one or more data records, each data record comprising data and meta data;
• in real-time, apply one or more tokenizer objects to a received data record, wherein the processor executes the machine instructions to:
• identify one or more data elements in data of a data record and in meta data of the data record, and
• extract and buffer the one or more data elements;
• in real-time, apply each of the one or more buffered data elements to a feature detect tree, the feature detect tree comprising a root node and one or more subsequent branch nodes, a set of the one or more subsequent branch nodes comprising a path beginning at the root node and culminating in a leaf node, wherein the processor executes the machine instructions to:
• apply a data element to the root node of the feature detect tree and testing the data element to reject or first select the data element,
• apply a first-selected data element to the one or more subsequent branch nodes of one or more paths of the feature detect tree, and conduct at each of the one or more subsequent branch nodes, a test to reject or further select the first-selected data element, and
• stop, for a selected path, the applying and testing after a further-selected data element reaches the selected path's leaf node;
• and
• install, in real-time, each data element reaching the selected path's leaf node, as a feature, into one or more recommender algorithm models connected to the selected path's leaf node.

A real-time recommender modeling method, comprising:

• a computer executing instructions to generate a plurality of feature detection algorithms;
• the computer encapsulating one or more of the plurality of feature detection algorithms into a feature detection object using one or more Strategy/Composite design patterns to generate a plurality of Strategy/Composite feature detection objects, each of the plurality of Strategy/Composite feature detection objects comprising a plurality of Strategy design patterns;
• the computer generating a feature detect tree, the feature detect tree comprising a root node, a plurality of branch nodes, and a plurality of leaf nodes, the feature detect tree defined by a plurality of unique feature traversal paths beginning at the root node, moving to one or more subsequent branch nodes, and terminating at one of the leaf nodes;
• at each of the root node and the plurality of branch nodes, the computer installing one of more of the Strategy/Composite feature detection objects;
• using the Strategy/Composite feature detection objects, testing features to select or reject each of the features, comprising:
• applying a feature to the root node of the feature detect tree and conducting a first test of the feature to reject or select the feature,
• applying a feature selected at the root node to one or more subsequent branch nodes of one or more selected traversal paths of the feature detect tree, and conducting at each of the one or more subsequent branch nodes, a second test to reject or further select the feature, and
• stopping, for a selected traversal path, the applying and testing after the further selected feature reaches the selected traversal path's leaf node;
• the computer executing instructions to generate a plurality of feature install algorithms;
• the computer encapsulating each of the feature install algorithms in a feature installation object;
• the computer linking each of the leaf nodes to one of the feature installation objects;
• and
• the computer linking each of the feature installation objects to one of a plurality of recommender algorithm models.

A real-time recommender method of a RTRM system, comprising a processor:

• receiving and processing streaming data including data related to a new item and a user action from each of a first user and a second user;
• applying features extracted from the data to a feature detect mechanism, wherein the feature detect mechanism comprises a feature detect tree, the feature detect tree comprising a root node and one or more subsequent branch nodes, a set of the root node and one or more subsequent branch nodes comprising a path culminating in a leaf node;
• testing each applied feature using one or more feature detect algorithms to accept or reject each applied feature, comprising:
• applying a feature to the root node of the feature detect tree and conducting a first test of the feature to reject or select the feature,
• applying a feature selected at the root node to one or more subsequent branch nodes of one or more selected paths of the feature detect tree, and conducting at each of the one or more subsequent branch nodes, a second test to reject or further select the feature, and
• stopping, for a selected path, the applying and testing after the further selected feature reaches the selected path's leaf node;
• installing each accepted feature into one or more recommender algorithm models connected to the branch's leaf node of the feature detect tree;
• processing the user action and identifying the user;
• determining sufficient information already exists in the RTRM system for the first user;
• determining insufficient information already exists in the RTRM system for the second user;
• for the first user, employing the recommender algorithm models, using the information already existing in the RTRM system to generate and provide one or more recommendations to the first user;
• and
• for the second user, executing a cold start process employing at least one of the recommender algorithm models, using information included in the user action of the second user to generate and provide one or more recommendations to the second user.