Semantic based collaborative filtering

Details Semantic based collaborative filtering Semantic based collaborative filtering

1999-08-06

2002-11-26

Gravini, Stephen (Department: 3622)

Data processing: financial, business practice, management, or co

Automated electrical financial or business practice or...

Discount or incentive

Reexamination Certificate

active

06487539

ABSTRACT:

BACKGROUND
1. Technical Field
The present invention relates generally to an automated computer based method for making product recommendations over an electronic network. More particularly, the invention is directed to a method for making product recommendations using content and compatibility attributes. This approach is referred to herein as semantic collaborative filtering.
2. Background Description
With the recent increase in popularity of on-line shopping over the Internet, entities providing the shopping sites are interested in obtaining information about shoppers that would help in selling their products. Since it is often possible to track the shoppers at the various sites they visit, information concerning their buying habits may be ascertained and used by the shopping sites to provide personalized recommendations. To this end, several companies have become popular for providing techniques used to facilitate internet commerce.
For example, companies such as Likeminds, Inc. and Firefly Network, Inc. provide techniques for facilitating Internet commerce based on “collaborative filtering”. In collaborative filtering, recommendations are made to a potential purchaser based on past explicit ratings by other customers. These techniques are particularly useful in those cases in which the products are homogenous in nature. For example, in the case of Likeminds, customers are asked to provide ratings with respect to their preferences of a particular product such as compact discs. The ratings are based on the degrees of like and dislike a customer may have for the particular product in question. These ratings are then collected and archived for later use. At some point in the future a product recommendation will be made to a new customer based on the previously archived data of other customers.
However, the pure collaborative filtering approach does not work well if customers do not partake in the explicit rating of products. Unfortunately, customers in an e-commerce environment typically prefer to minimize their time on-line and, thus, are usually unwilling to spend extra time rating products.
Nonetheless, even if customers are willing to explicitly rate products, such ratings may be difficult to manage. Thus, in some circumstances implicit ratings may be desirable for use in making product recommendations. Implicit ratings refer to the set of products which have been bought or browsed by a customer.
Thus, as an alternative to collaborative filtering (which uses explicit ratings), some of the techniques used by companies to facilitate Internet commerce are based on implicit ratings. For example, Net Perceptions is a company which uses implicit ratings of products to provide product recommendations.
One technique for obtaining explicit ratings is commonly referred to as “content-based” filtering. Content based filtering uses extracted texts and other information from e-commerce websites to provide recommendations to potential purchasers.
An example of content-based filtering is the intelligent infrastructure offered by Autonomy, Inc. This system provides an Agentware content server, which is a scaleable content personalization and organization engine for Internet information providers. This technique extracts key concepts from documents and websites to automate the categorization, cross-referencing, hyperlinking, and presentation of the information. The customer profiling system of this software enables information and service providers to understand the interests of customers and deliver personalized information.
Another company which provides intelligent servers is Aptex Software. Aptex uses a form of content-based filtering referred to as “content mining”. In content mining, text and other unstructured content is automatically analyzed to make intelligent decisions and recommendations.
The use of content mining to provide product recommendations is described in U.S. Ser. No. 09/169,029, entitled “Content Based Method for Product-Peer Filtering, filed on Oct. 9, 1998, commonly assignee, and incorporated herein by reference. According to one aspect of that invention, a method for providing product recommendations to customers in an e-commerce environment, includes the steps of deriving product characterizations for each of said plurality of products. Individual customer characterizations are created on each of said customers based on usage of the product characterizations by each of the respective customers. Clustering is performed based on similarities in the customer characterizations, to form peer groups. Individual customers are categorized into one of the peer groups. Product recommendations are made to customers based on the customer characterizations and information from the categorized peer groups.
Despite the provision and availability of the above described recommendation techniques, a need still exists for a method which provides recommendations and affords substantial customer personalization to the product recommendation process, without resorting to explicit group product ratings.
SUMMARY OF THE INVENTION
The present invention is directed to a method for making product recommendations in an electronic commerce network.
According to a first aspect of the invention, there is provided a method for providing product recommendations to customers in an e-commerce environment. The method includes the step of generating content and compatibility representations of products corresponding to a plurality of customers. A similarity function is calculated between pairs of content attributes corresponding to the products. A similarity function is calculated between pairs of compatibility attributes corresponding to the products. The plurality of customers are clustered into a plurality of peer groups. For a given customer, a closest peer group of the plurality of peer groups is determined. At least one potential recommendation is then generated for the given customer based on the closest peer group.
According to a second aspect of the invention, the step of generating the content representations includes the steps of concatenating product descriptions of the products bought by an individual customer, and calculating a content vector from the concatenated product descriptions.
According to a third aspect of the invention, the step of generating the compatibility representations includes the steps of calculating a fraction of time that a categorical attribute takes on a given value, and calculating a compatibility vector from the calculated fraction of time.
According to a fourth aspect of the invention, the clustering step includes the step of calculating a similarity function between a pair of categorical values as a predefined function of a support of the pair of categorical values.
According to a fifth aspect of the invention, the clustering step includes the steps of calculating a content-similarity between the content vector of a particular customer and content centroids of the plurality of customers, and calculating a compatibility-similarity between the compatibility vector of the particular customer and compatibility centroids of the plurality of customers.
According to a sixth aspect of the present invention, the step of determining the closest peer group includes the steps of determining a predetermined number of closest clusters to the given customer, and designating a union of the closest clusters as the closest peer group.
According to a seventh aspect of the invention, the step of generating the at least one potential recommendation includes the steps of determining most frequently bought products by the customers in the closest peer group, and recommending at least one of the most frequently bought products to the given customer.
According to an eighth aspect of the invention, the method further includes the step of filtering the at least one potential recommendation using domain specific rules.
According to a ninth aspect of the invention, the filtering step includes the step of determining whether any of the domain specific rules are relevant to the given cust

Affiliated with

Also associated with

Rating

Semantic based collaborative filtering does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Semantic based collaborative filtering, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Semantic based collaborative filtering will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2967744