Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberUS20010049623 A1
Publication typeApplication
Application numberUS 09/169,029
Publication dateDec 6, 2001
Filing dateOct 9, 1998
Priority dateOct 9, 1998
Also published asUS6356879
Publication number09169029, 169029, US 2001/0049623 A1, US 2001/049623 A1, US 20010049623 A1, US 20010049623A1, US 2001049623 A1, US 2001049623A1, US-A1-20010049623, US-A1-2001049623, US2001/0049623A1, US2001/049623A1, US20010049623 A1, US20010049623A1, US2001049623 A1, US2001049623A1
InventorsCharu C. Aggarwal, Philip S. Yu
Original AssigneeCharu C. Aggarwal, Philip S. Yu
Export CitationBiBTeX, EndNote, RefMan
External Links: USPTO, USPTO Assignment, Espacenet
Content based method for product-peer filtering
US 20010049623 A1
Abstract
The present invention derives product characterizations for products offered at an e-commerce site based on the text descriptions of the products provided at the site. A customer characterization is generated for any customer browsing the e-commerce site. The characterizations include an aggregation of derived product characterizations associated with products bought and/or browsed by that customer. A peer group is formed by clustering customers having similar customer characterizations. Recommendations are then made to a customer based on the processed characterization and peer group data.
Images(5)
Previous page
Next page
Claims(22)
What is claimed is:
1. A method for providing product recommendations to customers in an e-commerce environment, comprising the steps of:
(a) deriving product characterizations for each of said plurality of products;
(b) creating individual customer characterizations on each of said customers based on usage of said product characterizations by each of said respective customers;
(c) clustering based on similarities in said customer characterizations, to form peer groups;
(d) categorizing individual customers into one of said peer groups; and
(e) making product recommendations to customers based on said customer characterizations and information from said categorized peer groups.
2. The method according to
claim 1
, wherein the step of creating customer characterizations includes extracting product characterizations when said customers browse or purchase said products.
3. The method according to claim 27 wherein the step of creating customer characterizations further includes concatenating each of the product characterizations of all products browsed or purchased by an individual customer.
4. The method according to
claim 1
, wherein the step of deriving product characterizations includes use of text characterizations associated with said plurality of products.
5. The method according to
claim 1
, wherein said step of creating product characterizations further includes the step of:
(a) finding a frequency of occurrence for each word in said text descriptions;
(b) dividing the total frequency for each word by the frequency of occurrence for said word for all customers;
(c) finding the standard deviation for each word;
(d) selecting words having larger standard deviations; and
(e) expressing a product characterization based on said selected words.
6. The method of
claim 1
, further including the steps of:
(a) deriving product characterizations based on a current on-line session;
(b) accessing historical product characterizations from memory;
(c) creating a customer characterization by weighted concatenating characterizations from steps (a) and (b);
(d) computing a cluster centroid for each of said peer groups;
(e) selecting the peer groups whose cluster centroid is closest to the characterization created in step (c); and
(f) generating one of product, peer and profile recommendations based on said selected peer group.
7. The method of
claim 6
, wherein said cluster centroid is computed by concatenating text characterizations of all customer characterizations in each peer group.
8. The method of
claim 6
, wherein the recommendations comprise a weighted concatenation of text-characterizations of products bought and browsed in said current on-line session.
9. A stored program device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform process steps for providing product recommendations to customers in an e-commerce environment, comprising the steps of:
accessing product information on products offered for sale on an e-commerce site;
deriving product characterizations from said product information;
accessing customer usage information relating to said products;
creating customer characterization on each of said customers based on said accessed usage information;
clustering based on similarities in said customer characterizations, to form a plurality of peer groups;
receiving queries from a current session customer;
creating present customer characterizations associated with said current session customer;
categorizing said current session customer into a selected peer group based on similarities in the present customer characterizations and past customer characterizations;
responding to said queries with product recommendations to said current session customer based at least on information from said categorized selected peer group.
10. The device according to
claim 9
, further including stored programs for causing said computer to extract data exchanged between said current session customer and said site to create present customer characterizations.
11. The device according to
claim 9
, wherein the step of creating customer present characterizations further includes weighted concatenating each of the product characterizations of all products browsed or purchased by said present session customer.
12. The device of
claim 9
, further including stored programs for causing said computer to:
compute a cluster centroid for each of said peer groups by concatenating text characterizations of all customer characterizations in each peer group;
select the peer group having a cluster centroid closest to the present customer characterizations; and
generate one of product, peer and profile recommendations based on data from said selected peer group.
13. In a computer having a processor and stored program for causing the computer to interact with customers in an e-commerce environment and to provide answers to customer inquiries, said stored program comprising:
means for deriving product characterizations for a plurality of products;
means for creating customer characterizations based on usage of said product characterizations by said customers;
means for clustering, based on similarities in said customer characterizations, for forming a plurality of peer groups;
means for placing individual customers into one of said peer groups; and
means for providing answers to said customer queries based on said customer characterizations and information from said peer groups.
14. The stored program of
claim 13
, wherein said means for creating customer characterizations includes storage for archiving said customer characterizations.
15. The stored program of
claim 13
, further including means for creating present customer characterizations for a currently on-line customer based on a concatenation of data extracted from usage of product characterizations by the currently on-line customer.
16. The stored program of
claim 15
, further including means for placing a currently on-line customer in one of said peer groups based on similarities in said present customer characterizations and stored characterizations.
17. The stored program of
claim 13
, wherein said means for deriving product characterizations include means for extracting text descriptions associated with said products.
18. The stored program of
claim 13
, wherein said means for providing answers includes means for recommending products to an individual customer based on product characterizations from products browsed by said individual customer in a current session.
19. The stored program of
claim 13
, wherein said means for providing answers includes means for recommending products to an individual customer based on a weighted frequency of historic product characterizations and a weighted frequency of present product characterizations from products browsed by said individual customer in a current session.
20. The stored program of
claim 13
, wherein said means for providing answers includes means for recommending products to an individual customer based on historic product characterizations.
21. The stored program of
claim 13
, wherein said means for providing answers includes means for recommending products to an individual customer based on products in a promotion list.
22. The stored program of
claim 13
, wherein said means for providing answers includes providing a peer distribution profile based on product frequency for one of said peer groups.
Description
    BACKGROUND OF THE INVENTION
  • [0001]
    1. Field of Invention
  • [0002]
    The present invention relates to an automated computer based apparatus and method for making product recommendations over an electronic commerce network. More particularly, the invention is directed to an apparatus and method for making product recommendations based on customer user behavior.
  • [0003]
    2. Discussion of Related Art
  • [0004]
    With the recent increase in popularity of on-line shopping over the Internet, entities providing the shopping sites are interested in obtaining information on shoppers that would help in selling their products. The traditional customer or market surveys used in obtaining such information are applicable and usable for the providers. Companies such as Likeminds, Inc. (www.likeminds.com) and Firefly Network, Inc. (www.firefly.com) provide survey information which are based on explicit ratings by customers, commonly referred to in the art as recommendation engines using ‘collaborative filtering’. The use of engines based on ratings have particular applicability in products which are uniform and of a particular type. For example, in the case of Likeminds, customers are asked to provide ratings for preferences for products such as compact discs, based on their degrees of likes or dislikes. 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. Recommendations made based on past explicit ratings by customers are known as ‘collaborative filtering’.
  • [0005]
    The collaborative filtering approach does not work well if the customers do not participate in the explicit ratings of products. A customer or purchaser in an e-commerce environment typically prefers to minimize his time on-line and is usually unwilling to spend extra time, especially in rating products, on-line or otherwise.
  • [0006]
    Other companies offer ‘content-based’ filtering which uses extracted texts and information from e-commerce websites.
  • [0007]
    An example is the intelligent infrastructure offered by Autonomy, Inc. (www.autonomy.com) . 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.
  • [0008]
    Another company which provides intelligent servers is Aptex Software (www.aptex.com). Aptex uses a ‘Content Mining’ method which automatically analyzes text and other unstructured content to make intelligent decisions and recommendations.
  • [0009]
    Net Perceptions is still another company (www.netperceptions.com) which uses (Grouplens) implicit or explicit ratings of products to provide recommendations. Implicit ratings refer to the set of products bought or browsed by a customer.
  • [0010]
    Despite the provision and availability of the above described intelligent servers, a need still exists for a method or system for facilitating characterization of customers and products on the basis of customers' natural browsing/purchasing behavior, without resorting to explicit group product ratings, and providing recommendations based on peer group categorization, affording substantial customer personalization to the product recommendation process.
  • SUMMARY OF THE INVENTION
  • [0011]
    It is therefore an object of the present invention to provide a system and a method of using characterizations of products and user behavior, including user browsing or purchasing behavior, to generate product recommendations at an e-commerce site.
  • [0012]
    It is another object of the present invention to utilize the characterizations to create peer groups to personalize the recommendation to users. A peer group is a collection of customers whose product preferences have been previously archived and whom display a pattern of product preferences similar to that of the new customer.
  • [0013]
    The above objectives are accomplished by a method according to the present invention, which provides product recommendations to customers in an e-commerce environment, comprising the steps of: deriving product characterizations from each of a plurality of products; creating individual customer characterizations on each of the customers based on usage of the product characterizations by each of the respective customers; clustering based on similarities in the customer characterizations, to form peer groups; categorizing individual customers into one of the peer groups; and making product recommendations to customers based on the customer characterizations and the categorized peer groups.
  • [0014]
    The step of creating customer characterizations preferably includes extracting product characterizations when the customers browse or purchase the products, and the step of creating customer characterizations may include concatenating each of the product characterizations of all products browsed or purchased by an individual customer.
  • [0015]
    Preferably, the product characterizations are derived from text characterizations of each of the products and may further include the steps of: finding the frequency of occurrence for each word in the text descriptions; dividing the total frequency for each word by the frequency of occurrence for the word for all customers; finding the standard deviation for each word; selecting words having larger standard deviations; and expressing a product characterization based on the selected word.
  • [0016]
    Another method of the invention preferably include the steps of: (a) deriving product characterizations based on a current on-line session; (b) accessing historical product characterizations from memory; (c) creating a customer characterization by weighted concatenating characterizations from steps (a) and (b); (d) computing a cluster centroid for each of the peer groups; (e) selecting the peer groups whose cluster centroid is closest to the characterization created in step (c); and (f) generating one of product, peer and profile recommendations based on the selected peer group. The cluster centroid may be computed by concatenating text characterizations of all customer characterizations in each peer group, and the recommendations may comprise a weighted concatenation of text-characterizations of products bought and browsed in the current on-line session.
  • [0017]
    A system according to the present invention provides, a computer having a processor and stored program for causing the computer to interact with customers in an e-commerce environment and to provide answers to customer inquiries, the stored program comprises: means for deriving product characterizations for a plurality of products, means for creating customer characterizations based on usage of the product characterizations by the customers, means for clustering, based on similarities in the customer characterizations, for forming a plurality of peer groups, means for placing individual customers into one of the peer groups, and means for providing answers to the customer queries based on the customer characterizations and information from the peer groups.
  • [0018]
    The system preferably includes storage for archiving the customer characterizations.
  • [0019]
    The stored program further includes means for creating present customer characterizations for a currently on-line customer based on a concatenation of data extracted from usage of product characterizations by the currently on-line customer.
  • [0020]
    The stored program further includes means for placing a currently on-line customer in one of the peer groups based on similarities in the present customer characterizations and stored characterizations.
  • [0021]
    The means for deriving product characterizations preferably includes means for extracting text descriptions associated with the products.
  • [0022]
    The means for providing answers preferably includes means for recommending products to an individual customer based on product characterizations from products browsed by the individual customer in a current session.
  • [0023]
    The means for providing answers preferably includes means for recommending products to an individual customer based on a weighted frequency of historic product characterizations and a weighted frequency of present product characterizations from products browsed by the individual customer in a current session.
  • [0024]
    The means for providing answers also preferably includes means for recommending products to an individual customer based on historic product characterizations.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • [0025]
    [0025]FIG. 1 is a block diagram of the illustrative system according to the present invention.
  • [0026]
    [0026]FIG. 2 is a flow diagram illustrating an example of an overall operation of the system according to the present invention.
  • [0027]
    [0027]FIG. 3 is a flow diagram illustrating a method of creating product characterizations according to the present invention.
  • [0028]
    [0028]FIG. 4 is a flow diagram illustrating a method for generating customer characterizations according to the present invention.
  • [0029]
    [0029]FIG. 5 is a flow diagram illustrating a method for generating recommendations according to the present invention.
  • DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
  • [0030]
    The present invention can be generally characterized as comprising two stages; a pre-processing stage followed by an on-line querying stage. The pre-processing stage generally comprises converting text descriptions of products available at an e-commerce site into concise product characterizations. For purposes of this disclosure, the term ‘characterization’ can be described as a set of words with appropriate weights. The weights are defined by the frequency of occurrence of the words in the text descriptions of the products. The characterizations can be used as indicators of customer behavior. For example, for an application in which movies are being recommended to a customer, the valuable product characterizations may refer to text describing the actors and actresses along with the class of the movie.
  • [0031]
    Once the product characterizations are created individual customer characterizations are then created. A customer characterization is created by concatenating each of the product characterizations of all products which were either bought or browsed by that customer. For purposes of the present invention, concatenations can be described as a process of adding the sets of words and weights to form longer characterizations. For example, given characterization 1: heavy (1), metal (1), SONY (2); and characterization 2: SONY (1) Music (3), the concatenation of the two characterizations gives the following characterization:
  • [0032]
    A Concatenated Characterization
  • [0033]
    heavy (1) metal (1) SONY (3), Music (3).
  • [0034]
    The numbers in brackets correspond to the weights.
  • [0035]
    Peer groups of customers can then be formed by clustering the customer characterizations; generally, customers exhibiting likeminded buying/browsing habits are clustered into a peer group. Once the clusters are generated the pre-processing stage is complete. The pre-processing stage is performed at prespecified timed intervals to generate the text characterizations and clusters. Clusters are used to respond to the queries posed at the on-line query stage.
  • [0036]
    The on-line query stage follows the pre-processing stage and generally comprises one or more customers making queries to a server over a network. The server in turn computes the results of the queries and returns them to each respective customer.
  • [0037]
    Referring now to the drawings, FIG. 1 shows a preferred architecture according to an illustrative embodiment of the present invention. A server 8 is connected to multiple customers or clients 4 over a network 6. The clients make requests (queries) 7 to server 8 and server 8 responds to the requests by returning a result 9 for each query made. The network 6 is an electronic commerce network capable of facilitating product sales and purchases. The network is preferably the Internet.
  • [0038]
    The server 8 comprises a CPU 10 for coordinating and controlling server 8 and performs the necessary computations for generating a query result 9. Main memory 14 acts as a data repository for each customer 4. The customer data may also reside on disk 15 or in cache 16. When a customer characterization is derived from historical data, that historical data is retrieved from the data repository for that customer.
  • [0039]
    Turning to FIG. 2, a flowchart illustrating an example of an overall operation of the present invention, having a pre-processing stage of steps 210 and 220, and an on-line querying stage of steps 230, 240 and 250.
  • [0040]
    At step 210, text descriptions associated with each product at the e-commerce site are converted into product characterizations. For purposes of the present disclosure, product characterizations are text characterizations which have been filtered to remove extraneous terminology (i.e. common language). The process of converting text descriptions into product characterizations generally includes selecting those words contained within the text description of each product offered at the e-commerce site which prove to be valuable indicators of customer behavior. For example, at a site at which movies or CDS are recommended, some examples of words with high inference power could be “action”, “romance”, “mystery”, “drama”, etc. This is because these words are highly indicative of customer behavior, and are likely to have considerable skew in their distribution across the different customers. The steps associated with creating product characterizations will be discussed in further detail below in connection with FIG. 3.
  • [0041]
    Next, at step 220, the product characterizations are used for grouping customers into clusters or peer groups. Peer groups constitute sets or clusters of customers whose purchasing/browsing characteristics closely match that of a customer for which a recommendation is to be made. Clustering algorithms are well known in the art. Preferably, the clustering method described in “System and Method for Detecting Clusters of Information,” C. Aggarwal et al., a commonly assigned patent application, Ser. No. 09/070,600, filed on Apr. 30, 1998, is used to carry out the clustering operation. The disclosure of that application is incorporated by reference herein.
  • [0042]
    The on-line querying stage begins at step 230, where queries from one or more customer are received for processing. The process of receiving customer requests can be considered substantially continuous in the sense that customers may issue requests to the server at anytime or at random intervals. A customer request is received at step 230. Based on the pre-processed data (i.e. formation of peer groups or clusters) generated at the pre-processing stage a response to the customer query is generated at step 240.
  • [0043]
    At step 250, a determination is made whether a predetermined time interval has elapsed since the pre-processing stage was last performed. If the time interval has not elapsed, the process remains in the 230-250 loop responding to whatever additional user queries may exist in the queue. Otherwise, when the time interval lapses the process returns to step 210 to perform the pre-processing functions. This timing structure facilitates handling of customer queries ‘on the fly’. Since the pre-processing stage is more time consuming as compared to the querying stage, the system is preferably set to perform pre-processing less frequently but whenever the on-line querying is experiencing less traffic.
  • [0044]
    Referring next to FIG. 3, an illustrative flowchart of creating product characterizations from text descriptions (step 210) according to one embodiment of the present invention is shown. At step 310, the process determines the average frequency with which each word was bought by each customer. For purposes of the present embodiment, a word is considered bought by a customer if it occurs in the text characterization for a product bought or browsed by that customer. For example, if the same product was bought k times, then the complimentary word is also counted k times and the total frequency for that word is k. Other embodiments may consider bought words as those texts or words browsed by a customer, or by predefined combinations of browsing and buying. Each word bought by a customer is counted and summed as the total frequency. At step 320, the total frequency counts are converted into a fraction, by dividing the counts for each customer by the total frequency with which each word was “bought” by all customers, defined as F(word, customer). At step 330, a standard deviation of the fractions of words which were bought by the different customers is computed. The standard deviation for the fractions of words bought by a given customer is computed as follows: St . Dev . ( word ) = { [ 1 N ( F ( word , customer ) - 1 / N ) 2 ] / N } 1 / 2
  • [0045]
    where N is the total number of customers
  • [0046]
    In general, words from a text description having low inference power (non-descriptive) will include generic words of the language (i.e. is, the, an) and will expectedly have low standard deviation values. In contrast, words which have high inference power, which might include, for example, words such as a singer=s name or a music category as part of a CD music product recommendation environment will expectedly have high standard deviation values. Those words with the highest standard deviations are selected to create a product characterization from an input text description of a product.
  • [0047]
    As an example, the original text description for a compact disc could be:
  • [0048]
    Name of CD: Whitford/St. Holmes-Whitford/St. Holmes
  • [0049]
    Label: SONY MUSIC
  • [0050]
    Genre: General
  • [0051]
    Original Release Date: 1981
  • [0052]
    Engineer: George Pappas
  • [0053]
    Producer: Tom Allom
  • [0054]
    No. of Discs: 1
  • [0055]
    Mono/Stereo: Stereo
  • [0056]
    Studio/Live: Studio
  • [0057]
    Category: Heavy Metal
  • [0058]
    After performing the procedure in Flowchart 3, the characterized words are: Heavy, Metal, General, SONY, Tom, Allom. This is because these words are highly indicative of the nature of the CD, while the remaining words are simply general words. Words like Date, Label, etc. are likely to occur in each and every text description, and therefore their distribution will not be skewed, as all people are likely to browse these words regardless of “taste”. On the other hand, for words like “heavy”, “metal” etc., they are likely to be highly skewed in favor of people who access these words frequently.
  • [0059]
    Reference is now made to FIG. 4, an illustrative flowchart describing the steps associated with step 220 of FIG. 2, that of generating clusters (peer groups) from the generated product characterizations. At step 410, a customer characterization is created by concatenating the product characterizations for all products that the customer either bought or browsed. Concatenation in the current context means that each of the individual product characterizations bought or browsed by a customer will be appended to form one long characterization. At step 420, each customer is clustered into peer groups based on the customer characterizations formed at the previous step. The peer groups define sets of customers who exhibit similar buying/browsing behavior. For example: consider product characterizations created by the following words: chip, hardware, PC, SoftwareGames, PCGames, Joystick, processor, Aptiva, Windows98 . . . etc. In a cluster of customers having three words ‘SoftwareGames, PCGames, and Joystick’ occurring repeatedly signal that this cluster of customers correspond to a group of people who are interested in computer games. Note that even though Joystick, PCGames and SoftwareGames are not names of individual products, but words characterizing products, the amount of information contained in this set of words as belonging to a cluster is behavior information and certain assumptions can be made about the people in that cluster.
  • [0060]
    Turning now to FIG. 5, a flowchart illustrating how the process responds to various queries corresponding to customer requests is provided (step 240, FIG. 2). FIG. 5 defines the on-line query stage. It is important to note that each of the process steps illustrated in FIG. 5 are described with regard to a single customer request. It can be appreciated, however, that the method can accommodate, on a first come first serve basis, a plurality of requests from one or more customers.
  • [0061]
    At step 502, a first input is received by the system which comprises all of the previously generated clusters or peer groups. Next at step 505, A second and third input, e.g., a second input buying behavior and a third input browsing behavior, are received by the process where the second input is defined as comprising product characterizations generated in a current on-line buying/browsing session. The product characterizations can be generated as specified in FIG. 3. Preferably, the generation is in real time while a customer browses/purchases products at a current on-line session. The third input, product characterizations are similar to the second input, but the product characterizations are retrieved from historical, archived data stored in main memory 14. The generated data includes product characterizations from previous browsing/purchasing on-line sessions specific to that customer. At step 510, a customer characterization is generated for that customer. Preferably, the customer characterization is constructed utilizing the first input and clusters in conjunction with either or both the second and third inputs. When both the second and third inputs are utilized they are combined by weighted concatenation. The weighted concatenation is derived by multiplying the frequency of each word from the first and second inputs by an appropriate weight. Taking the example given above for concatenation of the CD, wherein characterization A: heavy (1), metal (1), SONY (2); and characterization B: SONY (1) Music (3), if the concatenation A is multiplied by 2, and concatenation B is multiplied by 3, the weighted concatenation is the sum of that product, yielding: heavy (2), metal (2), SONY (7), and Music (9). The weights correspond to the relative importance of the buying and browsing behavior, respectively.
  • [0062]
    At step 520, a plurality of cluster centroids are generated. A single cluster centroid is computed for each peer group generated at step 420 of FIG. 4. The cluster centroids are generated by concatenating the text characterizations of all the individual customer characterizations in that cluster (peer group). At step 530, a closest cluster centroid is found to the customer characterization derived at step 510. A “closest” cluster centroid to a customer characterization can be achieved by objective functions known to one skilled in the art. Examples of objective functions usable for a “closeness” determination include cosine angle, the dice coefficient and the Jaccard coefficient. See “Information Retrieval, Data Structure and Algorithms”, by William B. Frakes, and Ricardo Baeza-Yates, [Prentice Hall, 1992], which the use of coefficients to define similarity (i.e. closeness) is described, the disclosure therein is incorporated herein by reference. At step 540, a list of peers are found constituting those customers who are members of the cluster (peer group) whose associated cluster centroid was found to be “closest” to the customer characterization. At step 550, the peer group is utilized to respond to the queries posed in the particular e-commerce environment.
  • [0063]
    Examples of utilized queries will be described in conjunction with FIG. 5. As a further note each of the following queries requires the first input defined at step 502, however, certain of the queries may use either one or both of the inputs defined at step 505. The exemplary queries are;
  • [0064]
    1. Query
  • [0065]
    For a set of products bought/browsed, find the list of all products which form the best recommendation list.
  • [0066]
    A weighted concatenation is performed on product characterizations of the products bought/browsed by the customer in the current session (the second input). The third input representing product characterizations from historical data is not used. A report of the closest peers is made at step 540. The products with the highest frequency bought by the closest peers are then reported as the recommendations.
  • [0067]
    2. Query
  • [0068]
    For a given customer, and a set of products browsed/bought by him, find the best recommendation list.
  • [0069]
    This query is similar to query 1 except that a weighted frequency of the third input at step 505 is now utilized along with a weighted frequency of the second input.
  • [0070]
    3. Query
  • [0071]
    For a given customer, find the best recommendation list.
  • [0072]
    This query is similar to the first and second, however, only the third input from step 505 is used. In effect, the second input is discarded.
  • [0073]
    4. Query
  • [0074]
    Find the recommendation list out of a prespecified promotion list for the queries (1),(2), and (3).
  • [0075]
    This query differs from the first three in that the product recommendation list would be filtered to exclude all products except those contained on a prespecified promotion list as specified at step 550.
  • [0076]
    5. Query
  • [0077]
    Find the profile of the customers who are interested in a particular product.
  • [0078]
    This query is implemented in a way similar to that described for query 1. It differs at step 550 of FIG. 5 in that a peer distribution profile is output based on the product frequency for that peer group. A peer distribution profile is a profile of the customers in the group, such as age, sex, etc. For example, age histograms can be output.
  • [0079]
    6. Query
  • [0080]
    For a given profile of customers, find the products that they would most like.
  • [0081]
    This query contemplates customer profile queries that may take the form: For people in the 20-30 age group, what products would they most like? This query is answered by concatenating the text characterizations of all customers within the specified profile set at step 510.
  • [0082]
    7. Query
  • [0083]
    For the queries (1),(2), and (3), generate the characterizations which would be most liked by a customer, rather than the products themselves.
  • [0084]
    For this query the cluster centroid, derived at step 520, provides the required product characterizations.
  • [0085]
    8. Query
  • [0086]
    Find the closest peers for a given customer.
  • [0087]
    For this query step 530 provides the required response. Namely, a report of the customers contained within the selected centroid.
  • [0088]
    Advantageously, the illustrative method and system according to the present invention facilitates characterization of customers and products on the basis of customers natural browsing/purchasing behavior, without resorting to group product ratings. The customer characterizations become the means by which product recommendations are eventually generated thereby affording substantial personalization to the product recommendation process.
  • [0089]
    Although illustrative embodiments of the present invention have been described herein with reference to the accompanying drawings, it is to be understood that the invention is not limited to those precise embodiments, and that various other changes and modifications may be affected therein by one skilled in the art without departing from the scope or spirit of the invention.
Referenced by
Citing PatentFiling datePublication dateApplicantTitle
US7685017May 16, 2001Mar 23, 2010Fujitsu LimitedCommunication intermediating method, intermediating device, and communication device
US7716224Jun 14, 2007May 11, 2010Amazon Technologies, Inc.Search and indexing on a user device
US7853900Dec 14, 2010Amazon Technologies, Inc.Animations
US7865817Mar 29, 2007Jan 4, 2011Amazon Technologies, Inc.Invariant referencing in digital works
US7921309Jun 14, 2007Apr 5, 2011Amazon TechnologiesSystems and methods for determining and managing the power remaining in a handheld electronic device
US8032506 *Aug 25, 2000Oct 4, 2011Andrej GregovUser-directed product recommendations
US8078502Feb 1, 2007Dec 13, 2011International Business Machines CorporationPopulating an E-commerce shopping cart and other E-commerce fields based upon content extracted from natural language input
US8131647Jan 19, 2005Mar 6, 2012Amazon Technologies, Inc.Method and system for providing annotations of a digital work
US8175989Jan 3, 2008May 8, 2012Choicestream, Inc.Music recommendation system using a personalized choice set
US8234282Jul 31, 2012Amazon Technologies, Inc.Managing status of search index generation
US8266173Sep 11, 2012Amazon Technologies, Inc.Search results generation and sorting
US8341210Jun 14, 2007Dec 25, 2012Amazon Technologies, Inc.Delivery of items for consumption by a user device
US8341513Jun 14, 2007Dec 25, 2012Amazon.Com Inc.Incremental updates of items
US8352449Mar 29, 2006Jan 8, 2013Amazon Technologies, Inc.Reader device content indexing
US8364516Jun 29, 2009Jan 29, 2013The Nielsen Company (Us), LlcMethods and apparatus to determine the effects of trade promotions on subsequent sales
US8378979Jan 27, 2009Feb 19, 2013Amazon Technologies, Inc.Electronic device with haptic feedback
US8380579Mar 1, 2011Feb 19, 2013Manyworlds, Inc.Contextual commerce systems and methods
US8417772Aug 10, 2011Apr 9, 2013Amazon Technologies, Inc.Method and system for transferring content from the web to mobile devices
US8423889Dec 11, 2008Apr 16, 2013Amazon Technologies, Inc.Device specific presentation control for electronic book reader devices
US8458119Aug 18, 2011Jun 4, 2013Manyworlds, Inc.People matching in subscription-based communities
US8458120Aug 18, 2011Jun 4, 2013Manyworlds, Inc.Search-based people matching system and method
US8478716Aug 18, 2011Jul 2, 2013Manyworlds, Inc.Proximity-based people matching system and method
US8515900Aug 18, 2011Aug 20, 2013Manyworlds, Inc.Environment-responsive people matching system and method
US8515901Nov 9, 2011Aug 20, 2013Manyworlds, Inc.Explanatory people matching system and method
US8554640 *Aug 19, 2010Oct 8, 2013Amazon Technologies, Inc.Content completion recommendations
US8566263Jun 15, 2008Oct 22, 2013World Assets Consulting Ag, LlcAdaptive computer-based personalities
US8571535Sep 14, 2012Oct 29, 2013Amazon Technologies, Inc.Method and system for a hosted mobile management service architecture
US8600920Mar 24, 2007Dec 3, 2013World Assets Consulting Ag, LlcAffinity propagation in adaptive network-based systems
US8606770Oct 3, 2011Dec 10, 2013Andrej GregovUser-directed product recommendations
US8656040Jun 14, 2007Feb 18, 2014Amazon Technologies, Inc.Providing user-supplied items to a user device
US8666844 *Jun 22, 2010Mar 4, 2014Johnson & Johnson Consumer CompaniesIntelligent performance-based product recommendation system
US8700005Jun 14, 2007Apr 15, 2014Amazon Technologies, Inc.Notification of a user device to perform an action
US8725565 *Sep 29, 2006May 13, 2014Amazon Technologies, Inc.Expedited acquisition of a digital item following a sample presentation of the item
US8793575Nov 11, 2011Jul 29, 2014Amazon Technologies, Inc.Progress indication for a digital work
US8832584Mar 31, 2009Sep 9, 2014Amazon Technologies, Inc.Questions on highlighted passages
US8843406Dec 27, 2007Sep 23, 2014Yahoo! Inc.Using product and social network data to improve online advertising
US8898283 *Apr 9, 2010Nov 25, 2014Alibaba Group Holding LimitedRecommendation of network object information to user
US8954444Apr 14, 2010Feb 10, 2015Amazon Technologies, Inc.Search and indexing on a user device
US8965807Jun 14, 2007Feb 24, 2015Amazon Technologies, Inc.Selecting and providing items in a media consumption system
US8990215Jun 14, 2007Mar 24, 2015Amazon Technologies, Inc.Obtaining and verifying search indices
US9087032Jan 26, 2009Jul 21, 2015Amazon Technologies, Inc.Aggregation of highlights
US9116657Nov 18, 2010Aug 25, 2015Amazon Technologies, Inc.Invariant referencing in digital works
US9129289 *Sep 29, 2006Sep 8, 2015Drew VaughnSystems and methods for providing remote ordering capabilities
US9158741Oct 28, 2011Oct 13, 2015Amazon Technologies, Inc.Indicators for navigating digital works
US9178744Dec 20, 2012Nov 3, 2015Amazon Technologies, Inc.Delivery of items for consumption by a user device
US9208437Dec 13, 2012Dec 8, 2015Alibaba Group Holding LimitedPersonalized information pushing method and device
US9219797Oct 3, 2011Dec 22, 2015Amazon Technologies, Inc.Method and system for a hosted mobile management service architecture
US20020083119 *May 16, 2001Jun 27, 2002Kazuki MatsuiCommunication intermediating method, intermediating device, and communication device
US20040172267 *Aug 19, 2003Sep 2, 2004Jayendu PatelStatistical personalized recommendation system
US20060161578 *Jan 19, 2005Jul 20, 2006Siegel Hilliard BMethod and system for providing annotations of a digital work
US20060235721 *Jun 13, 2006Oct 19, 2006Lawrence KavanaghSystem and method for managing marketing applications for a website
US20060259344 *Jul 18, 2006Nov 16, 2006Choicestream, A Delaware CorporationStatistical personalized recommendation system
US20060294084 *Jun 28, 2006Dec 28, 2006Patel Jayendu SMethods and apparatus for a statistical system for targeting advertisements
US20070088624 *Sep 29, 2006Apr 19, 2007Drew VaughnSystems and methods for providing remote ordering capabilities
US20070112637 *Nov 13, 2006May 17, 2007The Procter & Gamble CompanyInteractive peer validation for product choices
US20070203589 *Nov 13, 2006Aug 30, 2007Manyworlds, Inc.Adaptive Recombinant Process Methods
US20070255707 *Apr 25, 2007Nov 1, 2007Data Relation LtdSystem and method to work with multiple pair-wise related entities
US20080183554 *Jan 29, 2007Jul 31, 2008W.W. Grainger, Inc.System and method for providing customized catalogs
US20080189187 *Feb 1, 2007Aug 7, 2008International Business Machines CorporationPopulating an e-commerce shopping cart and other e-commerce fields based upon content extracted from natural language input
US20080288354 *Jul 24, 2008Nov 20, 2008Manyworlds Inc.Location-Aware Adaptive Advertising
US20080295039 *Jun 14, 2007Nov 27, 2008Laurent An Minh NguyenAnimations
US20090171748 *Dec 27, 2007Jul 2, 2009Yahoo! Inc.Using product and social network data to improve online advertising
US20100063870 *Mar 11, 2010Anderson Gregory DMethods and apparatus to determine the effects of trade promotions on competitive stores
US20100070871 *Mar 18, 2010International Business Machines CorporationExtendable Recommender Framework for Web-Based Systems
US20100169162 *Jun 29, 2009Jul 1, 2010Anderson Gregory DMethods and apparatus to determine the effects of trade promotions on subsequent sales
US20100169174 *Jun 29, 2009Jul 1, 2010Anderson Gregory DMethods and apparatus to determine the effects of trade promotions on competitive products
US20100262556 *Oct 14, 2010Johnson & Johnson Consumer Companies, Inc.Intelligent performance-based product recommendation
US20100262692 *Oct 14, 2010Alibaba Group Holding LimitedRecommendation of network object information to user
US20100301114 *May 19, 2010Dec 2, 2010Lo Faro Walter FMethod and system for transaction based profiling of customers within a merchant network
US20110066497 *Aug 30, 2010Mar 17, 2011Choicestream, Inc.Personalized advertising and recommendation
US20110093341 *Apr 21, 2011Manyworlds, Inc.Explanatory Advertising Systems and Methods
US20110153452 *Jun 23, 2011Manyworlds, Inc.Contextual Commerce Systems and Methods
US20120030029 *Feb 2, 2012Manyworlds, Inc.System and Method for Adaptive Videos
US20120030063 *Feb 2, 2012Manyworlds, Inc.Physical Site-Contextualized Recommender
US20130097053 *Sep 20, 2011Apr 18, 2013University-Industry Cooperation Group Of Kyung- Hee UniversityMethod and system for recommending a combined service by taking into account situation information on a target user and the degree of complementarity of a service
US20130198022 *Oct 18, 2011Aug 1, 2013Alibaba Group Holding LimitedMethod and Apparatus of Determining A Linked List of Candidate Products
US20150051974 *Aug 27, 2014Feb 19, 2015Alibaba Group Holding LimitedRecommendation of network object information to user
USRE43768Oct 24, 2011Oct 23, 2012Manyworlds, Inc.Adaptive commerce systems and methods
USRE44559Oct 19, 2011Oct 22, 2013World Assets Consulting Ag, LlcAdaptive social computing methods
USRE44966Oct 28, 2011Jun 24, 2014World Assets Consulting Ag, LlcAdaptive recommendations systems
USRE44967Oct 28, 2011Jun 24, 2014World Assets Consulting Ag, LlcAdaptive social and process network systems
USRE44968Oct 28, 2011Jun 24, 2014World Assets Consulting Ag, LlcAdaptive self-modifying and recombinant systems
USRE45770Dec 26, 2012Oct 20, 2015World Assets Consulting Ag, LlcAdaptive recommendation explanations
WO2007127296A2 *Apr 25, 2007Nov 8, 2007Philippe AnkaouaSystem and method to work with multiple pair-wise related entities
WO2009085524A2 *Nov 26, 2008Jul 9, 2009Yahoo IncUsing product and social network data to improve online advertising
Classifications
U.S. Classification705/14.51
International ClassificationG06Q30/02, G06Q30/06, G06F17/30, G06F13/00
Cooperative ClassificationY10S707/99933, Y10S707/99943, G06Q30/0253, G06Q30/0631, G06Q30/0625, G06Q30/02
European ClassificationG06Q30/02, G06Q30/0631, G06Q30/0253, G06Q30/0625
Legal Events
DateCodeEventDescription
Oct 9, 1998ASAssignment
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:AGGARWAL, CHARU C.;YU, PHILIP S.;REEL/FRAME:009517/0954
Effective date: 19981008
Jul 7, 2005FPAYFee payment
Year of fee payment: 4
Jul 17, 2009FPAYFee payment
Year of fee payment: 8
Dec 27, 2012ASAssignment
Owner name: EBAY INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:INTERNATIONAL BUSINESS MACHINES CORPORATION;REEL/FRAME:029536/0508
Effective date: 20120928
Aug 14, 2013FPAYFee payment
Year of fee payment: 12