Collaborative-filtering contextual model optimized for an objective function for recommending items
View Patent ↗Methods and apparatus for a recommendation system based on collaborative filtering is provided. Explicit and implicit ratings of items by network users are used to create a contextual model. The explicit ratings comprise different rating types regarding different item attributes. The implicit ratings comprise different rating types derived from different user events and may include recency, intensity, or frequency ratings. The contextual model may be optimized for a specific objective function, such as click-through-rate or conversion rate. In other embodiments, item information is used to produce a content model where item information for an item is encoded as metadata into a document that represents the item. The contextual or content model is used to recommend one or more items to a current user. The basic unit of the recommendation system may be an item set of two or more items or a particular sequence of two or more items.
1. A method for implementing a collaborative-filtering based recommendation system for recommending one or more items among a plurality of items to a current user of a network, an item representing a product, service, webpage, audio, or document, the method comprising:
providing computer system hardware for performing:
producing a model comprising a plurality of similarity measurements, each similarity measurement reflecting a level of similarity between two items in the plurality of items, wherein the model is produced using ratings of the plurality of items from a plurality of previous network users, the ratings comprising a plurality of different rating types, the model being optimized for maximizing a click-through-rate or conversion rate of the one or more recommended items by using an overall rating equation for determining an overall rating of each item in the plurality of items by each previous network user, wherein the overall rating equation comprises a predetermined weight value for each rating type, the weight values for the plurality of different rating types being predetermined so as to maximize the click-through-rate or conversion rate of the one or more recommended items, a rating type that is more predictive of click-through-rate or conversion rate having a higher predetermined weight value than a rating type that is less predictive of click-through-rate or conversion rate, the overall rating of an item comprising a weighted sum of a plurality of different rating types;
receiving a first rating of a first item from the current user; and
determining the one or more recommended items by producing a predicated rating for each item in the plurality of items, the predicated rating of an item being produced using the received first rating and a similarity measurement, retrieved from the model, that reflects a level of similarity between the item and the first item.
2. The method of claim 1 , further comprising:
serving information relating to the one or more recommended items to the current user, wherein:
information relating to the one or more recommended items comprises text, an image, a hyperlinik, video, or audio.
3. The method of claim 1 , wherein:
the click-through-rate of an item relates to the rate at which network users click on an item given a number of times the item is viewed by the network users; and
the conversion rate of an item relates to the rate at which a particular action regarding an item is received from network users given a number of times the item is viewed by the network users.
4. The method of claim 3 , wherein a particular action regarding an item comprises purchasing the item, subscribing to the item, filling out a survey regarding the item, or downloading the item.
5. The method of claim 1 , wherein:
the ratings from the plurality of previous network users comprises a plurality of different explicit and implicit ratings;
explicit ratings comprise a plurality of different explicit rating types for a plurality of different item attributes; and
implicit ratings comprise a plurality of different implicit rating types based on a plurality of different user events comprising a plurality of different user interactions with a server.
6. The method of claim 1 , wherein an item further comprises a set of two or more items or a specific sequence of two or more items.
7. The method of claim 1 , wherein the plurality of similarity measurements comprises a plurality of different levels of similarity.
8. The method of claim 1 , wherein the predicated rating of an item comprises the rating predicted to be given by the current user for the item.
9. The method of claim 1 , wherein determining the one or more recommended items further comprises selecting the items having the top Y predicated ratings, Y comprising a predetermined number.
10. A system for implementing a collaborative-filtering based recommendation system for recommending one or more items among a plurality of items to a current user of a network, an item representing a product, service, webpage, audio, or document, the system comprising:
a server computer system comprising:
a scoring module configured for:
producing a model comprising a plurality of similarity measurements, each similarity measurement reflecting a level of similarity between two items in the plurality of items, wherein the model is produced using ratings of the plurality of items from a plurality of previous network users, the ratings comprising a plurality of different rating types, the model being optimized for maximizing a click-through-rate or conversion rate of the one or more recommended items by using an overall rating equation for determining an overall rating of each item in the plurality of items by each previous network user, wherein the overall rating equation comprises a predetermined weight value for each rating type, the weight values for the plurality of different rating types being predetermined so as to maximize the click-through-rate or conversion rate of the one or more recommended items, a rating type that is more predictive of click-through-rate or conversion rate having a higher predetermined weight value than a rating type that is less predictive of click-through-rate or conversion rate, the overall rating of an item comprising a weighted sum of a plurality of different rating types;
receiving one a first rating of a first item from the current user; and
determining the one or more recommended items by producing a predicated rating for each item in the plurality of items, the predicated rating of an item being produced using the received first rating and a similarity measurement, retrieved from the model, that reflects a level of similarity between the item and the first item.
11. The system of claim 10 , wherein:
the scoring module is further configured for sewing information relating to the one or more recommended items to the current user; and
information relating to the one or more recommended items comprises text, an image, a hyperlinik, video, or audio.
12. The system of claim 10 , wherein:
the click-through-rate of an item relates to the rate at which network users click on an item given a number of times the item is viewed by the network users; and
the conversion rate of an item relates to the rate at which a particular action regarding an item is received from network users given a number of times the item is viewed by the network users.
13. The system of claim 12 , wherein a particular action regarding an item comprises purchasing the item, subscribing to the item, filling out a survey regarding the item, or downloading the item.
14. The system of claim 10 , wherein:
the ratings from the plurality of previous network users comprises a plurality of different explicit and implicit ratings;
explicit ratings comprise a plurality of different explicit rating types for a plurality of different item attributes; and
implicit ratings comprise a plurality of different implicit rating types based on a plurality of different user events comprising a plurality of different user interactions with a server.
15. The system of claim 10 , wherein an item further comprises a set of two or more items or a specific sequence of two or more items.
16. The system of claim 10 , wherein the plurality of similarity measurements comprises a plurality of different levels of similarity.
17. The system of claim 10 , wherein the predicated rating of an item comprises the rating predicted to be given by the current user for the item.
18. The system of claim 10 , wherein the scoring module is further configured for determining the one or more recommended items by selecting the items having the top Y predicated ratings, Y comprising a predetermined number.