TWI635450B - Personalized product recommendation method - Google Patents

Abstract

The invention relates to a personalized product recommendation method, the steps of which mainly include constructing a plurality of positioning sensors in a sales field, and each of the positioning sensors divides each area of the sales field to establish a spatial index value, and then Retrieving information in the sales field through a recommendation module carried by the consumer, which can return the signal strength between the recommended module and the sensor or query the product information on the Internet, and then collect it through a recommendation server. Calculate information such as the consumer's moving line, dwell time, distance from the product classification, and give the consumer a matching behavior label, and then instantly recommend the goods around the consumer that match their behavioral label or The classification of goods to consumers is a recommended method of considering the potential purchase intention of consumers.

Description

 Personalized product recommendation method  

The invention relates to a digital product recommendation method, in particular to a personalized product recommendation method for collecting potential purchase intentions of consumers.

In the modern era of e-commerce, with the prevalence of virtual advertisements such as webpage advertisement ordering and online shopping malls, the willingness of consumers to store in physical stores has been gradually reduced. Among them, there is a big reason because the operating mode of physical storefronts Compared with the automatic statistical analysis and other mechanisms of the virtual store, it is more difficult to actually understand the contours of the customers and the products they are interested in, so it is difficult to provide incentives for consumers to browse the products in the physical store, or to purchase products with higher purchase rates for individual consumers. Planning a marketing strategy.

In the past, the product recommendation methods in the physical store mostly used the sales terminal records actually purchased by the consumers, and then the correlation results were calculated through the association rules or the collaborative filtering algorithm. However, such prior art only considered When the actual sales volume is reached, it does not take into account the consumers’ repeated hangover, hesitation before purchase, or online search for product-related information, which may show interest in the goods. These actions can effectively show the potential purchase intention of the consumers, but Rarely used for analysis, may lead to some unpopular but the opportunity for a particular consumer to purchase a product or product classification is recommended.

In view of the above-mentioned lack of prior art, the inventors of the present invention have considered and researched a recommendation method that can be based on the consumer's potential purchase intention, and can be used to collect consumers from the physical storefront or surrounding areas through the popularization of modern social personal mobile devices. Behavior, and conversion to understanding the information that consumers may be interested in to make recommendations, is an effective recommendation method for individuals.

The present invention proposes a personalization product recommendation method, which is a method for labeling consumers and recommending products in consideration of potential purchase intentions of consumers, that is, to supplement the unfocused parts of the prior art.

The personalized product recommendation method of the present invention is applied to a sales field, such as a department store or a sales circle, in which a plurality of positioning sensors are disposed in the sales field, and each of the positioning sensors can be WiFi. Or a Beacon device or the like for providing customers with online browsing information in the sales field, and each of the positioning sensors can be used as a positioning reference or a coordinate in the sales field; the method of the present invention will be based on each The location of the location sensor divides the sales field into regions, and establishes a spatial index value for each region of the sales field, and displays the products to be sold in each region within the sales field. In summary, according to the planning of the sales field, the spatial index value of each area may correspond to at least one item classification.

Then, since the personal mobile device is already quite popular, the consumer terminal of the method of the present invention is installed by installing a recommendation module on the mobile device that the consumer carries, and the consumer will carry the recommended module. The mobile device moves, stays or accesses the information in the sales field. The recommendation module mainly returns the recommendation module itself to receive the sensing signals sent by the positioning sensors in the sales field. The intensity change data is transmitted to a recommendation server of the present invention, and the recommendation server calculates the stay position, time and movement route of the consumer carrying the recommendation module in the sales field, and generates according to the Regarding a first field attribute attribute data of the consumer in the sales field; wherein each of the positioning sensors of the present invention can perform triangulation, indoor positioning based on signal RSSI intensity, and quadtree (Quadtree) Or a technique such as R-Tree to locate the location of the recommendation module in the sales field.

In addition, in the method of the present invention, since a plurality of the positioning sensors can provide a network connection function, when the mobile device is installed by the consumer, the mobile device and the network function are installed. When the positioning sensor performs network connection, each of the positioning sensors can further obtain the behavior of the consumer inquiring or browsing the webpage through the mobile device, and transmit to the recommendation server to generate a The second field of behavioral attribute data, this step is mainly to collect consumers through the Internet to query or browse basic information about products, other purchase channels, competitive prices, etc., to understand the consumer's domain for the sales field. Whether the goods in the product may be of interest or have a price comparison behavior.

Then, the recommendation server obtains the change of the relative position and the spatial index value of the recommended module and the displayed item in the sales field according to the behavior attribute data in the first field, and converts the value to the consumer to the sales field. The recommended purchase rate of the products displayed in each area of the domain, the recommendation server mainly receives the sensing signal strength of each positioning sensor along the time line according to the recommendation module included in the behavior attribute data in the first field. And continuously consider the distance range distribution that can be received at a certain point on the timeline under the condition of fixed signal strength, in order to comprehensively calculate the distance between the consumer and the commodity, in order to further define the possible purchase rate of the consumer for each commodity and the commodity category, wherein The higher the distance between the sensing signal and the stronger the signal strength, the higher the value of the possible purchase rate of the defined consumer.

In addition, the recommendation server further calculates the behavior attribute data of the second field through a string similarity algorithm to filter and compare the product information of the consumer through the mobile device to query or browse the webpage, for example, Comparing the product information contained in the webpage with a long browsing time of each consumer or the surrounding information about the merchandise to know whether the consumer is interested in certain merchandise or performing online comparison; wherein the string of the present invention Similarity algorithm can use Levenshtein Distance algorithm, N-gram algorithm, JaroWinkler algorithm, Soundex algorithm, Morris-Pratt algorithm, Knuth-Morris-Pratt algorithm, Aho-Corasick algorithm, Baker-Bird algorithm Techniques such as Gusfield's algorithm, Boyer-Moore algorithm or Wu-Manber algorithm are implemented.

Then, the recommendation server will analyze the possible purchase rate according to the behavior attribute data in the first field and the effect of the webpage product comparison result of the behavior attribute data in the second field on the possible purchase rate. Purchasing the probability data for a consumer; in addition, the recommendation server is further connected with the sales terminal of each area in the sales field to obtain the actual number of consumers in the sales field in the sales area. The purchase of the product information, and consolidate to adjust the consumer purchase probability data, so that the consumer purchase probability data covers the potential purchase intention of the consumer as well as the actual purchase behavior.

Subsequently, the recommendation server introduces an external definition data, which is mainly related to the sales field sales strategy or sales time factor, etc., such as the floor distribution of the sales field, brand name, brand value, promotion schedule, Seasonal, sales time or government-scheduled holiday schedules, etc., the recommendation server filters the information in the consumer behavior attribute data according to the externally defined data or the factors actively deployed by the server manager, thereby giving The consumer who carries the recommendation module has a behavioral label, and after repeatedly performing the foregoing collecting and labeling steps, the consumers in the sales field domain can be divided into a plurality of groups according to their respective behavior labels, and Corresponding to each group, the recommendation server establishes a product classification profile for which the behavior tag has a corresponding consumption tendency, for example, a consumer who is satisfied with a product with a high discount rate in the sales field, or accesses the Internet. Consumers who inquire about the price of goods with high discount rates may be given the behavior tag of the discount family and be recommended by the servo. The device is divided into groups of discount families, or may be subdivided into groups according to different levels of discount rates, and the like.

The recommendation server may use the DBSCAN (Density-based Spatial Clustering of Applications with Noise), the OPTICS (Ordering Points To Identify the Clustering Structure), and the grouping method of the same type of the group according to the behavior label. Technical implementations such as K-Means, Hierarchical Agglomerative Clustering, Hierarchical Divisive Clustering, STING (Statistical Information Grid), CLIQUE (Clustering High-Dimensional Space), Wave-Cluster, COBWEB, CLASSIT or SOM (Self-Organizing Maps).

Finally, the recommendation server of the present invention generates a recommendation list according to a recommendation algorithm, and dynamically obtains a spatial index value of the area where the recommendation module is currently located in the sales field, and then places the recommendation list in the recommendation. The product information around the module is instantly transmitted to the recommendation module of each consumer to recommend the product information conforming to the consumer behavior label to the consumer; wherein the recommendation algorithm is Apriori, FP using the association rule algorithm. -growth, Eclat, etc., or AprioriSome, DynamicSome, FreeSpan, PrefixSpan, SPADE, SPIRIT, CloSpan, etc. for sequential correlation algorithms, or User-based, Item-based, Model-based, Content-Based for collaborative filtering algorithms Collaborative Filtering, etc., or Spark and other technologies for massive data analysis.

In this way, the product information recommended by the personalized product recommendation method of the present invention can comprehensively consider various substantive or potential purchase intentions of the consumer, and more in line with the consumer's expectation than the prior art recommendation method.

S100~S400‧‧‧Step procedure

S100~S104‧‧‧Step procedure

S201~S205‧‧‧Step procedure

S301~S303‧‧‧Step procedure

S401~S404‧‧‧Step procedure

1 is a flow chart showing the main steps of the personalized product recommendation method of the present invention.

2 is a flow chart of the first sub-step of the personalized product recommendation method of the present invention.

3 is a flow chart of a second sub-step of the personalized product recommendation method of the present invention.

4 is a flow chart showing a third sub-step of the personalized product recommendation method of the present invention.

Figure 5 is a flow chart showing the fourth sub-step of the personalized product recommendation method of the present invention.

The present invention will be further described in conjunction with the following embodiments. The background of the present embodiment is that a company intends to use the personalized product recommendation method of the present invention in a department store of the company to perform the user entering the department store. Products Featured.

Please refer to FIG. 1 , which is a flowchart of the main steps of the personalized product recommendation method of the present invention, which is a step of constructing a positioning sensor in step S100 to obtain consumer behavior data, and then screening external statistical filtering for step S200. The consumer is tagged and grouped, and the recommendation list is generated for the recommendation algorithm in step S300, and finally the dynamic recommendation is performed in step S400; wherein the detailed sub-step flow of each main step will be described in the following paragraphs.

Please refer to FIG. 2 , which is a flow chart of the first sub-step of the personalized product recommendation method of the present invention, which is used to explain the sub-step flow chart of the step S100 in FIG. 1 , wherein the step S100 includes four sub-steps, respectively. For S101~S104, sub-step S101 is to establish a spatial index. First, in order to obtain consumer behavior data, it is necessary to selectively construct a plurality of positioning sensors on each floor in the department store in order to The positioning sensor is used to establish the coordinates of each position in the department store, and the correlation information of the positioning sensor is correlated with the commodity position and the commodity classification of each area in the department store through a recommendation server to summarize Which positioning sensors are distributed in the same product classification area, and then the plane coordinates of each positioning sensor are clustered, and the lower left corner and the upper right corner of the range in which the positioning sensors are clustered are calculated. Positioning with a central coordinate, etc., to plan a geographical area of a rectangle, the recommendation server establishes a spatial index by the geographic extent of the different commodity classifications, that is, Each geographical area in the cargo company has a spatial index value. Therefore, it also means that each sensor in the geographical area has a spatial index value, and these positioning sensor codes, commodity classification information, and spatial index values are used. An area index correspondence table is created, and an example of the area index correspondence table can be shown in the following table.

Next, the sub-step S102 is performed to collect and analyze the first intra-domain behavior attribute data of the consumer, wherein the consumer installs a recommendation module about the department store on the mobile device owned by the consumer, when the recommendation module is When the user moves or stays in the department store, the recommendation module transmits the sensing signal data sent by the positioning sensors in the current geographical area to the recommendation server, and the recommendation server further The triangulation technology or the indoor positioning technology based on the strength of the RSSI of the signal is used to locate the plurality of positioning sensors that are closest to the consumer in unit time, and then the sensor is associated with the regional index correspondence table. The device can know that the recommendation module of the consumer is within or around the geographic category of the product category, that is, obtaining the first intra-domain behavior attribute data of the consumer; Further, based on information such as signal strength, distance or dwell time, the distribution operation can be summarized as a possible purchase by the consumer to purchase this product category. Rate, which rate may be purchased in the following table with reference to Fan.

Then, the sub-step S103 collects and analyzes the behavioral attribute data of the second field of the consumer, that is, collects the consumption in the same unit time by collecting the unit time of the behavioral attribute data of the first domain in the consumer. The mobile device browses the data of the webpage through the location sensor provided by the mobile device, which is the behavior attribute data of the second field, and the behavioral data of the second field is analyzed and analyzed. Knowing whether the content of the webpage browsed by the consumer in the geographical area includes the product webpage or the competition webpage on the Internet, and correlates the webpage content with the merchandise classification displayed in the department store or the information of the merchandise itself, Adjust the consumer's possible purchase rate.

Next, it is sub-step S104, that is, to continuously obtain the first field intra-domain behavior attribute data of the consumer and the second field intra-domain behavior attribute data, and analyze the possible purchase of each item by the consumer in each domain. Rate to generate a consumer purchase probability data.

Following the sub-step S104, it is the external statistical screening of the step S200 of the present invention. For the flowchart of the sub-steps to which it belongs, please refer to FIG. 3, wherein the step S200 includes five sub-steps; first, the sub-step S201 will consume The purchasing probability data of the person is statistically analyzed by introducing externally defined data. For example, whether the department store company in the present embodiment has a certain discounted product in the current marketing strategy, or a discount A degree of external data or the like is introduced into the recommendation server for statistical analysis to generate consumer quantity distribution data.

Then, the sub-step S202 is dynamically drawn to map the number of consumers, and the recommendation server will continuously collect the consumer behaviors represented by all the recommended modules in the department store, and dynamically analyze the statistics continuously; and the decision makers of the department store According to the information of the quantity distribution of the consumer, the sub-step S203 is configured to set the screening condition value; then, the sub-step S204 is grouped by the customer group, and the recommendation server will filter out the matching customer group from the consumer according to the screening condition value, and The customer group sets the behavior tag. Finally, the sub-step S205 is performed to establish the product classification profile, that is, the recommendation server aggregates the customer group of the same behavior tag and establishes the product classification profile of interest, for example, in this embodiment. If the imported external definition data does have discounts for certain active product categories, etc., it can analyze some consumers who are satisfied or satisfied with the discount of the goods. The recommendation server will establish a consumer for these consumers. The behavioral label of the discount family, and the consumer purchase rate information of these discount family consumers, to establish a discount family for them A classification contour correspondence table of possible commodities may be, for example, one of the contours of the discount family's interest classification may be a commodity with a discount rate of 50%.

Referring to FIG. 4, which is a third sub-step flow chart of the personalized product recommendation method of the present invention, to explain the recommendation algorithm of step S300 of the present invention, which includes three sub-steps; first, obtained in sub-step S301 The actual sales data, which is the recommended server to obtain the actual product sales information of the sales terminal in the department store, and associates the sold product classification with the location in the department store, for example, the consumer can The possible purchase rate of the actual purchased product classification is set to 1; then, the sub-step S302 is performed to adjust the consumer purchase rate, and the recommendation server associates the consumer purchase probability data generated in the above sub-step S104 with the result of the sub-step S301. The set operation is performed to adjust the consumer purchase probability data; then the sub-step S303 is used to perform the operation using the recommendation algorithm, and finally a recommendation list is generated.

Finally, please refer to FIG. 5 , which is a fourth sub-step flow chart of the personalized product recommendation method of the present invention, for explaining the dynamic recommendation of step S400 of the present invention, which also includes three sub-steps; first, tracking the consumption for sub-step S401 When the recommendation module of the consumer continues to receive the sensing signals of the plurality of positioning sensors in the department store, the user can locate the current position of the consumer and return the spatial index value of the area to The server is recommended; then, the sub-step S402 is a behavior tag query, and the recommendation server queries the consumer to belong to the discount family according to the consumer behavior continuously generated by the recommendation module or the identity code of the recommended module; The sub-step S403 is to pair the consumer peripheral products, wherein the recommendation server finds the corresponding product classification information according to the spatial index value returned by the recommendation module, and confirms whether the recommendation list generated after the sub-step S303 is met; Finally, it is recommended to the consumer for step S404, which is to filter the recommendation list near the location according to the consumer's discount family behavior label. Commodity classification for immediate recommendation to consumers.

After the present invention has been described in the above embodiments and drawings, it should be understood that the reader understands how the personalized product recommendation method of the present invention is implemented. However, it should be understood that the above description is for explanation and not for limitation. The scope of protection of the present invention.

In summary, the present invention is innovative in terms of technical ideas, and also has various functions that are not in the prior art, and has fully complied with the statutory invention patent requirements of novelty and progressiveness, and has filed a patent application according to law, and invites you to approve the invention. The patent application was inspired to invent, and it was a matter of feeling.

Claims (8)

A personalized product recommendation method, the method comprising: arranging a plurality of positioning sensors in a sales field, and dividing the sales field by the positioning sensors for clustering to sell the sales field in a recommendation server Each area of the field and the products displayed in the area establish a spatial index value; the recommendation module that is carried by the consumer feeds back information in the sales field, and the recommendation module returns the recommendation module to receive each location. The intensity change data of the sensing signal sent by the sensor is transmitted to the recommendation server, and the recommendation server calculates the stay position, time and movement route of the recommendation module in the sales field to generate information about the consumer. a first intra-domain behavior attribute data; providing a consumer mobile device connection network in each of the positioning sensors, thereby obtaining a behavior of the consumer inquiring or browsing the webpage by the mobile device, and transmitting to the recommendation servo And generating a second intra-domain behavior attribute data about the consumer; the recommendation server acquiring the recommendation module and the sales according to the first intra-domain behavior attribute data The relative position of the displayed items in the domain and the change of the spatial index value are converted into the possible purchase rate of the products displayed by the consumer in each area of the sales field, and the recommendation server calculates the second by a string similarity algorithm. Field behavior attribute data, in order to compare the consumer's mobile device query or browse the products included in the webpage, thereby adjusting the consumer's possible purchase rate, and then collecting it into a consumer purchase probability data; the recommendation server is introduced An externally defined information about the sales field sales strategy or sales time factor, the statistical filtering of the information in the consumer purchasing probability data, and the labeling of the consumer who carries the recommendation module, and then the consumption According to the behavior tag, it is divided into a group of the same type. And determining, by the group, a product classification profile corresponding to the consumption tendency of the behavior tag; the recommendation server calculating the consumer purchase probability data according to a recommendation algorithm to generate a recommendation list; and the recommendation server dynamically according to the The recommendation module is currently located in the spatial index value of the area in the sales field, and transmits the product information in the recommendation list that meets the outline of the product classification and is located around the recommendation module to the recommendation module for immediate recommendation to the consumer. The personalization product recommendation method of claim 1, wherein the recommendation server is further connected with a sales terminal of each area in the sales field to obtain a consumer who carries the recommendation module. Information on the actual purchase of goods in each area of the sales field to adjust the purchase probability data of the consumer. The personalization product recommendation method of claim 1, wherein the recommendation server receives the sensing of each of the positioning sensors according to the recommendation module included in the first field domain behavior attribute data. Signal strength, and consider the distance range distribution that can be received under this signal strength to calculate the distance between the consumer and the product, and then define the possible purchase rate of the consumer, wherein the stronger the sensing signal and the signal strength The closer the distance is, the higher the defined possible purchase rate value. The method for recommending personalized products as described in claim 1, wherein the externally defined data includes a floor distribution, a brand name, a brand value, a promotion schedule, a seasonality, a sales time, or a holiday schedule of the sales field. Cheng et al. The method for recommending personalized products according to claim 1, wherein each of the positioning sensors is transmitted through a triangular position and is based on a strong RSSI signal. Techniques such as indoor positioning, quadtree or R-Tree determine the location of the recommended module within the sales field. The method for recommending personalized products as described in claim 1, wherein the string similarity algorithm uses a Levenshtein Distance algorithm, an N-gram algorithm, a JaroWinkler algorithm, a Soundex algorithm, and a Morris-Pratt algorithm. Method, Knuth-Morris-Pratt algorithm, Aho-Corasick algorithm, Baker-Bird algorithm, Gusfield's algorithm, Boyer-Moore algorithm or Wu-Manber algorithm. The method for recommending personalized products according to claim 1, wherein the recommendation server uses a DBSCAN (Density-based Spatial Clustering of) for the grouping method in which the consumer is classified into the same type according to the behavior label. Applications with Noise), OPTICS (Ordering Points To Identify the Clustering Structure), K-Means, Hierarchical Agglomerative Clustering, Hierarchical Divisive Clustering, STING (Statistical Information Grid), CLIQUE (Clustering High-Dimensional Space), Wave-Cluster, COBWEB, CLASSIT or SOM (Self-Organizing Maps), etc. The personalization product recommendation method according to claim 1, wherein the recommendation algorithm used by the recommendation server is Apriori, FP-growth, Eclat, or the like, or a sequential association using an association rule algorithm. Algorithms such as AprioriSome, DynamicSome, FreeSpan, PrefixSpan, SPADE, SPIRIT, CloSpan, etc., or User-based, Item-based, Model-based, Content-Based Collaborative Filtering, or a huge amount of data analysis Spark and other technologies.