CN106779074B

CN106779074B - Market brand combination prediction method and prediction server

Info

Publication number: CN106779074B
Application number: CN201710053928.XA
Authority: CN
Inventors: 冯博; 张夏天; 王泽铭
Original assignee: Tengyun Tianyu Science & Technology Beijing Co ltd
Current assignee: Tengyun Tianyu Science & Technology Beijing Co ltd
Priority date: 2017-01-22
Filing date: 2017-01-22
Publication date: 2020-09-01
Anticipated expiration: 2037-01-22
Also published as: CN106779074A

Abstract

The invention discloses a market brand combination prediction method and a prediction server, wherein the method comprises the following steps: acquiring user geographical position information and brand information, and processing the user geographical position information and the brand information to acquire user characteristic information and first brand characteristic information; acquiring basic demographic data, and processing brand information, user characteristic information and the basic demographic data to acquire second brand characteristic information; respectively constructing a first deep learning neural network and a second deep learning neural network; taking the user characteristic information, the first brand characteristic information and the second brand characteristic information as input of a first deep learning neural network, and training the first deep learning neural network to obtain user purchase probability of each brand; and taking the user characteristic information, the second brand characteristic information and the user purchase probability as the input of a second deep learning neural network, and training the second deep learning neural network to predict the brand combination with the maximum user purchase probability.

Description

Market brand combination prediction method and prediction server

Technical Field

The invention relates to the technical field of big data processing, in particular to a market brand combination prediction method and a prediction server.

Background

With the continuous development of mobile internet, more and more people begin to use mobile terminal devices such as smart phones and tablet computers. Meanwhile, the widespread popularization of the mobile internet also promotes the development of mobile applications more rapidly, and when a user uses the installed mobile application on a mobile terminal, a series of state data, such as application information, mobile device information, environment information, location information, etc., is generated. The use of a large number of mobile devices generates massive data, and data representation can be realized by processing the massive data. In the data presentation, the user image (namely the characteristic distribution of the user) is extracted, and particularly the user image in a certain specific fence area has important significance for retailers, house producers and consumers. For example, by counting user figures in a certain market, the manager of the market can analyze the gender and age composition of the consumer, thereby realizing more accurate and targeted sales.

However, in the existing method, a large amount of online mall indoor positioning data and online mall sales activities are not perfectly combined, and accurate and effective analysis and prediction processing on mall customer group characteristics and mall brand combinations is difficult to realize.

Disclosure of Invention

To this end, the present invention provides a solution for market brand portfolio prediction in an effort to solve, or at least alleviate, the above-identified problems.

According to one aspect of the present invention, there is provided a shopping mall brand combination prediction method adapted to be executed in a prediction server, the prediction server including a data storage device, the data storage device storing therein user geographical location information of each user holding a mobile terminal in a shopping mall, brand information of each brand, and basic demographic data, the user geographical location information including a user identifier, a longitude and latitude, a time stamp, and a floor, the brand information including a brand name, brand geographical location information, brand commodity information, and brand selling information, the method including the steps of: acquiring user geographical position information and brand information from data storage equipment, and processing the user geographical position information and the brand information to acquire user characteristic information and first brand characteristic information, wherein the user characteristic information comprises user purchasing behavior and walking routes in a shopping mall; acquiring basic demographic data from a data storage device, and processing brand information, user characteristic information and the basic demographic data to acquire second brand characteristic information; respectively constructing a first deep learning neural network and a second deep learning neural network; taking the user characteristic information, the first brand characteristic information and the second brand characteristic information as input of a first deep learning neural network and the user purchasing behavior as labels for supervised learning, and training the first deep learning neural network to obtain the user purchasing probability of each brand; and training the second deep learning neural network by taking the user characteristic information, the second brand characteristic information and the user purchase probability as the input of the second deep learning neural network and taking a walking route in a shopping mall as a label for supervised learning so as to predict the brand combination with the maximum user purchase probability.

Optionally, in the shopping mall brand combination predicting method according to the present invention, the brand commodity information includes a brand flag under commodity name, a commodity price, a commodity quantity, a commodity shelf time, a commodity expected shelf off time, and whether season is due.

Optionally, in the shopping mall brand combination predicting method according to the present invention, the brand selling information includes a brand under-flag commodity name, a commodity selling price, a commodity selling quantity, and a user identification.

Optionally, in the shopping mall brand combination prediction method according to the present invention, the basic demographic data includes age, gender, income level, and industry of the department.

Optionally, in the shopping mall brand combination predicting method according to the present invention, the acquiring the user characteristic information and the first brand characteristic information includes: processing the brand name, the brand geographical position information and the user geographical position information to obtain the total residence time of the brand, the residence time of a single brand and a walking route in a mall; processing brand selling information, and acquiring user characteristic information by combining the residence time of a single brand and a walking route in a mall; and processing the total residence time of the brand and the walking route in the market to acquire the characteristic information of the first brand.

Optionally, in the shopping mall brand combination prediction method according to the present invention, the user characteristic information further includes: average purchase price of the goods, individual brand dwell time, and a frequent set of purchased brands.

Optionally, in the method for predicting a market brand combination according to the present invention, the first brand feature information includes an average brand staying time, a ratio of total brand staying times, and an optimal market walking route, where the optimal market walking route is the first K highest in repetition rate in the market walking route, and K is an integer not less than 1.

Optionally, in the shopping mall brand combination predicting method according to the present invention, the second brand feature information includes: the method comprises the following steps of average commodity price under a brand flag, brand popularity, brand target user information, brand geographical location information, average commodity price sold by the brand, the highest commodity quantity sold by the brand and the lowest commodity quantity sold by the brand.

Optionally, in the shopping mall brand combination prediction method according to the present invention, the method further includes: and respectively carrying out weighting processing on the user characteristic information and the first brand characteristic information according to the specific event.

Optionally, in the shopping mall brand combination prediction method according to the present invention, the method further includes: and coding the user characteristic information, the first brand characteristic information and the second brand characteristic information.

Optionally, in the shopping mall brand combination prediction method according to the present invention, the method further includes: acquiring geographical position information of the mobile terminals of each mobile terminal; connecting the market geographic position coordinates to form a market geofence; and filtering the geographical position information of the mobile terminal through a market geo-fence, and taking the geographical position information of the mobile terminal obtained after filtering as the geographical position information of each user holding the mobile terminal in the market.

According to yet another aspect of the present invention, a prediction server is provided that includes a data storage device, a first processing module, a second processing module, a construction module, a first training module, and a second training module. The data storage device stores user geographic position information of each user holding the mobile terminal in a shopping mall, brand information of each brand and basic demographic data, wherein the user geographic position information comprises user identification, longitude and latitude, timestamps and floors, and the brand information comprises brand names, brand geographic position information, brand commodity information and brand selling information; the first processing module is suitable for acquiring user geographic position information and brand information from the data storage device and processing the user geographic position information and the brand information to acquire user characteristic information and first brand characteristic information, wherein the user characteristic information comprises user purchasing behaviors and walking routes in a shopping mall; the second processing module is suitable for acquiring basic demographic data from the data storage device and processing the brand information, the user characteristic information and the basic demographic data to acquire second brand characteristic information; the building module is suitable for building a first deep learning neural network and a second deep learning neural network respectively; the first training module is suitable for taking the user characteristic information, the first brand characteristic information and the second brand characteristic information as input of a first deep learning neural network and taking the user purchasing behavior as a label for supervised learning, and training the first deep learning neural network to obtain the user purchasing probability of each brand; the second training module is suitable for taking the user characteristic information, the second brand characteristic information and the user purchase probability as input of a second deep learning neural network and taking a walking path in a shopping mall as a label for supervised learning, and training the second deep learning neural network so as to predict a brand combination with the maximum user purchase probability.

According to yet another aspect of the invention, there is also provided a computing device comprising at least one processor and at least one memory including computer program instructions, the at least one memory and the computer program instructions configured to, with the at least one processor, cause the computing device to perform a mall brand combination prediction method according to the invention.

According to the technical scheme of market brand combination prediction, user geographic position information and brand information are processed firstly to obtain user characteristic information and first brand characteristic information, then the brand information, the user characteristic information and basic demographic data are processed to obtain second brand characteristic information, then the user characteristic information, the first brand characteristic information and the second brand characteristic information are input into a first deep learning network to be trained to obtain user purchase probability of each brand, and finally the user characteristic information, the second brand characteristic information and the user purchase probability are input into a second deep learning network to be trained to predict a brand combination with the maximum user purchase probability. In the technical scheme, the user geographic position information, the brand information and the basic demographic data are processed to obtain the relevant characteristic information of the user and the brand, the characteristic information is further processed through weighting processing, coding processing and the like to improve the effectiveness of the characteristic information, the constructed first deep learning network is utilized to analyze the characteristic information of the user so as to obtain accurate user images of a shopping mall customer group, and finally, a brand and market combination is predicted through the constructed second deep learning network, so that the market is guided to plan a more efficient and reasonable brand combination to meet the user requirements, further more accurate oriented marketing activities are provided, and the perfect combination of a large amount of indoor positioning data accumulated on line and offline marketing activities of the shopping mall is realized.

Drawings

To the accomplishment of the foregoing and related ends, certain illustrative aspects are described herein in connection with the following description and the annexed drawings, which are indicative of various ways in which the principles disclosed herein may be practiced, and all aspects and equivalents thereof are intended to be within the scope of the claimed subject matter. The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description read in conjunction with the accompanying drawings. Throughout this disclosure, like reference numerals generally refer to like parts or elements.

FIG. 1 shows a flow diagram of a mall brand combination prediction method 100 according to one embodiment of the present invention;

FIG. 2 shows a schematic diagram of a prediction server 200 according to one embodiment of the invention;

FIG. 3 shows a schematic diagram of a prediction server 300 according to yet another embodiment of the invention;

FIG. 4 shows a schematic diagram of a prediction server 400 according to yet another embodiment of the invention; and

fig. 5 fig. 3 shows a schematic diagram of a prediction server 500 according to yet another embodiment of the invention.

Detailed Description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

FIG. 1 shows a flow diagram of a mall brand combination prediction method 100 according to one embodiment of the present invention. Mall brand combination forecasting method 100 is adapted to be executed in a forecasting server that includes a data storage device having stored therein user geographic location information, brand information for each brand, and basic demographic data for each user holding a mobile terminal within the mall. The shopping mall brand combination prediction method 100 obtains feature information related to the user and the brand by processing the various data, and trains and constructs a deep learning neural network by using the feature information to predict the brand combination with the maximum user purchase probability.

As shown in fig. 1, the method 100 begins at step S110. In step S110, first, user geographical location information and brand information are obtained from the data storage device, where the user geographical location information includes a user identifier, longitude and latitude, a timestamp, and a floor, and the brand information includes a brand name, brand geographical location information, brand commodity information, and brand selling information. Table 1 shows an example of storing user geographical location information of 1 user according to an embodiment of the present invention, which is specifically as follows:

TABLE 1

Geographical location information of the user identified as a1 when the user is active in mall S in 2016 12 months is shown in table 1, and for convenience of description, the user identified as a1 is denoted as user a 1. According to this embodiment, the number of users active in mall S in 2016 12 months totals 50000, with the user identifications of each user being A1, A2, A3, … …, A50000, respectively. As shown in table 1, the user identifier of the user is a1, which is used to uniquely determine and identify the user, the longitude and latitude are B1-B260, which respectively represent the longitude and latitude of the user a1 under the corresponding timestamp, each longitude and latitude includes a longitude and a latitude value, the floor represents the floor number of the floor where the user a1 is located at the current timestamp, and the shopping mall S is 5 floors in total. From table 1, it can be seen that, when the user a1 is located on floor 2, the latitudes and longitudes corresponding to the timestamps C1-C15 are kept unchanged at B1, which indicates that the user stays at the position corresponding to the latitude B1 in the floor 2 of the mall S in the time period corresponding to the timestamps C1-C15. Of course, user geographical location information for other users is also stored in the data storage device as in the storage example of table 1 above.

Table 2 shows an example of brand information storage of 1 brand according to an embodiment of the present invention, where the brand commodity information in the brand information includes a brand flag under-brand commodity name, a commodity price, a commodity quantity, a commodity shelf life, a commodity expected shelf life, and whether the commodity should be taken out of season, and the brand selling information includes a brand flag under-brand commodity name, a commodity selling price, a commodity selling quantity, and a user identifier, where units related to prices are all in terms of elements, and the following are specifically shown:

TABLE 2

Brand information of a brand having a brand name of P1 in the mall S is shown in table 2, and for convenience of description, the brand having a brand name of P1 is denoted by a brand name of P1. The geographical brand position information of the brand P1 is F2-03, which indicates that the shop corresponding to the brand is the 3 nd shop on the 2 nd floor in the mall S, and as can be seen from the brand commodity information, the brand flags have 50 commodity names in total, which are N1, N2, … … and N50 respectively. For convenience of description, the commercial product N1 is denoted by the trade name N1, and so on, taking the commercial product with the trade name N1 as an example. The price of the commodity N1 is 20 yuan, the quantity of the commodity is 15, the commodity shelf life is 2016, 9, 12 and predicted commodity shelf life is 2017, 2, 3 and is a season commodity. As can be seen from the brand selling information, the selling prices of the products N1-N50 are not completely consistent with the product prices, for example, the selling price of the product N3 is 32 yuan, and the product price is 36 yuan, which indicates that the product may be a corresponding product for a sales promotion such as a discount when sold. The user identifiers corresponding to the commodity N3 are all A20, and the number of sold commodities is 4, which indicates that 4 commodities N3 are sold to the same user A20. The user identifiers corresponding to the product N1 are a1 and a2, respectively, which indicate that the product is sold to different 2-bit users, the product selling price corresponding to the user identifier a1 is 20 yuan, the product selling quantity is 1, the product selling price corresponding to the user identifier a2 is 18 yuan, and the product selling quantity is 1, which indicates that the user a1 purchases 1 product N1, and the price at the time of purchase is 20 yuan, and the user a2 also purchases 1 product N1, but the price at the time of purchase is 18 yuan. It can also be seen from table 2 that, if the sales volume of the product N50 is 0, the sales price and the user identifier of the product are both null.

According to this embodiment, there are 101 brands in the mall S, the corresponding brand names P1, P2, … …, and P101, and the brand information of other brands is also stored in the data storage device as the storage example in table 2 above. It should be noted that the user geographic location information is generated after being processed based on positioning data acquired by Wi-Fi probe devices installed on each floor of a shopping mall, and a specific processing method is not shown here for the moment, and is explained later. The brand information of each brand is generally directly provided by a shopping mall, but the brand geographical position information can also be obtained by utilizing the Wi-Fi probe device, namely the relative position of each brand is collected in the process of collecting data by the Wi-Fi probe device, and the number of the Wi-Fi probe device represents the brand geographical position information.

In step S110, after the geographical location information and the brand information of the user are obtained from the data storage device, the geographical location information and the brand information of the user are processed to obtain user characteristic information and first brand characteristic information, where the user characteristic information includes a user purchasing behavior and a walking route in a shopping mall.

According to one embodiment of the present invention, the user characteristic information and the first brand characteristic information may be acquired in the following manner. Firstly, processing brand names, brand geographical location information and user geographical location information to obtain total residence time of brands, residence time of single brands and walking routes in a mall. Taking brand P1 in table 2 as an example, the brand geographical location information is F2-03, i.e., No. 2, store, there is a geofence corresponding to the area of the store, and when the user's longitude and latitude process the geofence, it indicates that the user stays at the brand. For user A1, when the user is at floor 2, the longitude and latitude corresponding to the timestamps C1-C15 are kept unchanged at B1, and the longitude and latitude B1 just falls into the geo-fence of the brand with the brand name P1, which indicates that the user stays at the brand with the stay time of C15-C1 in the time period corresponding to the timestamps C1-C15, and the stay time of the single brand of the user A1 is expressed as 'A1: T1', wherein the value of T1C 15-C1 is converted into a numerical value in minutes. According to the calculation method, the stay time lengths of the user A1 at the other brands are obtained on the one hand, and the stay time lengths of the other users at the brand P1 are counted and overlapped on the other hand, so that the total stay time length of the brand P1 is finally obtained. For other users and other brands in the shopping mall S, the single brand dwell time of each user in each brand and the total brand dwell time of all users in each brand are obtained by the processing method.

To obtain the route of the user a1, the brand that the user has left in the store is obtained. In the step of calculating the stay time of each user in each brand, it has been described that whether the user stays at a certain brand has been performed, and details thereof are not repeated herein. Considering that the walking route in the mall describes the route of the user moving in the mall in one day, the timestamps in the user geographical location information need to be segmented and divided according to the business hours of the mall every day, the timestamp segmentation sequence corresponding to each day is finally obtained, and the walking route in the mall on each day is obtained based on each timestamp segmentation sequence. Taking table 1 as an example, the timestamp sequences C1-C500 are composed of timestamps recorded by the activity of the user a1 in the mall S in 12 months of 2016, the business time of the mall S in 12 months is 10: 00-22: 00, namely from 10 am to 22 pm, the business time of each day is converted into corresponding daily business timestamps in combination with the date of the day, and the timestamp sequences C1-C500 are segmented according to the daily business timestamps, so as to obtain timestamp segmentation sequences of each day of the activity of the user a1 in the mall S in 12 months, which are respectively C1-C89, C90-C187, C188-C260, and C261-C500, and sequentially correspond to 3 days, 15 days, 18 days, and 24 days of 2016. This result indicates that user a1 has performed activity in mall S on 4 days of 2016, 12 months, 3 days, 15 days, 18 days, and 24 days. Based on the timestamp segmentation sequence, the corresponding longitude and latitude, the corresponding floor and the brand geographical position information of each brand, the brands where the user A1 respectively stays in the above 4 days are obtained, and finally the walking route of the user A1 in the mall within 12 months is obtained. Taking day 3 of 12 months as an example, brand names of brands which users stop one by one in the day are P1, P5, P16, P24 and P12, so that a walking line of the user A1 in a shopping mall S in day 3 of 12 months is formed and is expressed as' A1: [ P1, P5, P16, P24, P12 ].

And then, processing the brand selling information, and acquiring user characteristic information by combining the residence time of a single brand and a walking route in a store. According to this embodiment, the user purchasing behavior in the user characteristic information includes an average price and quantity of purchased brand under-flag goods. For a certain user, firstly, the commodity name, the commodity selling price and the commodity selling quantity corresponding to the user identification of the user are obtained from the brand selling information of a certain brand purchased by the user. For example, referring to table 2, user a1 stays at brand P1 and purchases a commodity, commodity names N1 and N2 corresponding to user identifier a1 are obtained from brand selling information of brand P1, the commodity selling price of commodity N1 is 20 yuan, the commodity selling quantity is 1, the commodity selling price of commodity N2 is 30 yuan, and the commodity selling quantity is 1. Then, the total price and total number of the commodities of the same brand purchased by the user are accumulated, and for the user a1, the total price of the commodities of the brand P1 purchased is 20+ 30-50 yuan, and the total number of the commodities is 1+ 1-2. Finally, the total number of the commodities is taken as the number of purchased under-brand flag commodities, the total commodity price is divided by the total number of the commodities as the average price of purchased under-brand flag commodities, the average price of the commodities under brand P1 purchased by the user a1 is 50/2-25 yuan, the number of purchased under brand P1 is 2, and the user purchasing behavior of the user a1 is expressed as "a 1: P1,25, 2". And for other users in the market S, acquiring the user purchasing behavior of each user by using the processing method.

According to yet another embodiment of the present invention, the user characteristic information further includes an average purchase price of the item, a stay time of the individual brand, and a frequent set of purchased brands. In this embodiment, the average purchase price of the product can be obtained in the following manner. For a certain user, obtaining the commodity name, the commodity selling price and the commodity selling quantity corresponding to the user identification of the user from the brand selling information of all brands purchased by the user, accumulating the total commodity price and the total commodity number of all commodities purchased by the user, and dividing the total commodity price by the total commodity number to be used as the average commodity purchasing price. For example, for user a1, the user purchased products of brand P1 and brand P15, commodity names N1 and N2 corresponding to user identifier a1 are obtained from brand selling information of brand P1, the commodity selling price of commodity N1 is 20 yuan, the commodity selling quantity is 1, the commodity selling price of commodity N2 is 30 yuan, the commodity selling quantity is 1, and the commodity name Q1 corresponding to user identifier a1 is obtained from brand selling information of brand P15, the commodity selling price of commodity Q1 is 70 yuan, and the commodity selling quantity is 3. The total price of all the commodities purchased by the user is accumulated, the total price of the commodities is 20+30+70 × 3 ═ 260 yuan, the total number of the commodities is 1+1+3 ═ 5, the total price of the commodities is divided by the total number of the commodities to be used as the average purchase price of the commodities, the value of the average purchase price is 260/5 ═ 52 yuan, and the average purchase price of the commodities of the user a1 is expressed as "a 1: 52". For other users in the market S, the average purchase price of the commodities of each user is obtained by using the processing method.

The method for acquiring the stay time of a single brand has been described above, and the frequent item sets of the purchased brands can be calculated according to Apriori algorithm or other algorithms of frequent cameras, all of which are easily conceivable to those skilled in the art of understanding the solution of the present invention and are within the scope of the present invention, and are not described herein again. For the representation method of the user purchasing the frequent item set of the brand, the user A1 is taken as an example for explanation, the frequent item set of the user purchasing the brand is represented as "[ P1, P15]: 0.3", [ P1, P15] represents the brand item set, and 0.3 represents the support degree.

In the above-mentioned case where brand information is typically provided by a store, but sometimes the store does not provide brand sales information in the brand information for a certain brand, then according to yet another embodiment of the present invention, the length of stay of a user at a single brand of that brand may be utilized to estimate whether there is payment activity during the stay. The estimation method comprises the following two steps, wherein one step is to preset a stay time threshold value, judge whether the stay time of a user in a single brand of a certain brand is larger than the stay time threshold value, if so, the user is shown to have purchasing behavior, and the other step is to perform regression prediction through brand information of other brands provided with brand selling information. If the user is judged to purchase a certain commodity of a certain brand according to the two methods, the user characteristic information is further acquired by combining the user geographical location information and the brand information of the brand.

And finally, processing the total residence time of the brand and the walking route in the market to obtain the characteristic information of the first brand. According to one embodiment of the invention, the first brand feature information comprises brand average stay time, brand total stay time ratio and an optimal market walking route, wherein the optimal market walking route is the first K market walking routes with the highest repetition rate, and K is an integer not less than 1. According to this embodiment, the brand average stay time may be obtained by, for each brand, counting the number of users who have stayed at the brand, and dividing the corresponding brand total stay time by the number of users who have stayed to obtain the brand average stay time for the brand, which is denoted as "P1: brand average stay time" in hours.

According to this embodiment, the total residence time ratio of the brand can be obtained in the following manner. After the total residence time of the brands is obtained, for each brand, the ratio of the total residence time of the brand to the total residence time of the brands of the brand to the left and right of the brand is obtained by combining the geographical position information of the brands, and the ratio is used as the ratio of the total residence time of the brand. For example, for a brand P1, the total residence time of the brand is 1000 hours, the brand geographic location of the brand P1 is F2-03, the brand with the brand geographic location information of F2-02 is the brand on the left of the brand P1, the corresponding total residence time of the brand is 1200 hours, the brand with the brand geographic location information of F2-04 is the brand on the right of the brand P1, the corresponding total residence time of the brand is 800 hours, the total residence time of the brand P1 is 1000/(1200+800) to 0.5, which is expressed as "P1: 0.5".

According to this embodiment, the optimal mall walking route can be obtained in the following manner. After the walking routes in the shopping malls of the users are obtained, the first K paths with the highest repetition rate are selected as the optimal shopping malls walking routes, and K is preferably 20. An example of an optimal mall walk route is given below, which is denoted as "[ P1, P24, P26, P30 ]", P1, P24, P26, and P30 are brand names, respectively.

Subsequently, step S120 is entered, and basic demographic data is first obtained from the data storage device, and table 3 shows an example of basic demographic data storage according to an embodiment of the present invention, where the basic demographic data includes age, gender, income level, and industry category, as follows:

age (year of old)	Sex	Income horizon (yuan)	The related industries
				25	1	4750	2
12	0	0	0
				45	1	7520	4
……	……	……	……
				70	0	2000	0

TABLE 3

Regarding the basic demographic data, the age and income level are integers not less than 0, the gender is male with 1, the gender is female with 0, the industry classification conforms to the national standard of national classification of national economy, the categories of all industries in the standard are A, B, … … and T which are 20 in total, for convenience of description, the categories A to T are sequentially corresponding to the categories 1 to 20, and the industry codes of the non-working population such as students are defined as 0. It can be seen that the first piece of data in table 3 corresponds to a male 25 years old, having an income level of 4750 yuan, who is engaged in category 2 industries, and so on.

Then, in step S120, the brand information, the user characteristic information, and the basic demographic data are processed to obtain second brand characteristic information, where the second brand characteristic information includes an average brand price under the brand flag, a brand popularity, brand target user information, brand geographical location information, an average brand price of sold goods, a maximum brand quantity of sold bid, and a minimum brand quantity of sold bid. The process of acquiring the geographical location information of the brand is described above, and is not described herein again.

According to one embodiment of the present invention, the average price of the brand-under-flag goods may be acquired in the following manner. And accumulating the prices and the quantity of the commodities under the brands according to the brand commodity information in the brand information of the brands to obtain the total price and the total quantity of the commodities under the brands, and calculating the average price of the commodities under the brands by dividing the total price of the commodities under the brands by the total quantity of the commodities under the brands. For example, for brand P1, the prices and quantities of 50 commodities under brand P1 are accumulated according to the data in table 2, the total price of the commodities under brand P1 is 27000 yuan, and the total quantity of the commodities under brand P1 is 750, so that the average price of the commodities under brand P1 is 27000/750-36 yuan, which is denoted as "P1: 36".

According to the embodiment, the brand popularity can be obtained by scoring according to the network search quantity, and the score range of the popularity score is 0-100. For example, there are 101 brands in total from P1 to P101 in the mall S, and the brands are arranged in the order of the smaller corresponding network search volume to the larger corresponding network search volume, and then the scores of 0 to 100 are sequentially obtained as the popularity scores of the brands. Brand P1 is at position 89 in the above arrangement, then a score of 88 is obtained as the popularity score for that brand, denoted as "P1: 88".

According to the embodiment, the brand target user information is obtained through processing of the user characteristic information and the basic demographic data. Table 3 gives an example of basic demographic data, and in fact each piece of data in table 3 is associated with a unique one of the user identifications. And the brand where each user stays can be known according to the walking route in the store of each user in the user characteristic information, so that the user where each brand stays and the user identification thereof can be obtained. For each brand, using the user identifier of the user where the brand stays, acquiring basic demographic data associated with the user identifier and processing the basic demographic data to obtain the average age of the user visiting the brand, the number of males and females in gender, the median value of income level and the industry distribution condition, and finally obtaining brand target user information, wherein the brand target user information comprises a brand name, a gender, an age, an income level and an industry code. For the brand target user information, the definition of gender and the industry code number is the same as that in the basic demographic data, and the ages are divided into 10 sections, and the numbers 0 to 9 are used as code numbers and respectively correspond to 10 age groups of 0 to 9, 10 to 19, 20 to 29, 30 to 39, 40 to 49, 50 to 59, 60 to 69, 70 to 79, 80 to 89 and 90 years and above. The income level in the brand target user information is divided into 9 grades, the grade 1-9 is represented by the number 1-9, and the grade 1-9 corresponds to the income of 9 stages, which are respectively 0 yuan, 0-2000 yuan, 2000-4000 yuan, 4000-6000 yuan, 6000-8000 yuan, 8000-10000 yuan, 10000-15000 yuan, 15000-20000 yuan and more. Taking brand P1 as an example, the user identifications of users who stayed at brand P1 are A1, A2, a13, a20, a105, a2056 and a13524, basic demographic data associated with the 7 users are obtained, and the average age of the users, the number of males and females in the statistical gender, the median of income levels and the industry distribution are calculated according to the basic demographic data, so that in users visiting brand P1, the average age of the users is 30 years, the number of males in the gender is 1, the number of females is 6, the median of income levels is 5421 yuan, the industry distribution is 5 people in 3 categories, 1 person in 1 category and 1 person in 7 categories. The average age of 30 years of the user falls into the 4 th interval of the age interval, the corresponding code is 3, the number of women in the gender is majority, therefore, the gender is represented by 0, the median 5421 yuan of income level falls into the 4 th grade of income level grade, the corresponding code is 4, the industry distribution situation is the most number of 3 types of people, the industry code is 3, and the finally obtained brand target user information of the brand P1 is as follows: brand name P1, gender 0, age 3, income level 4, and industry code 3, which are denoted as "P1: 0,3,4, 3".

According to this embodiment, the average price of the brand sold goods can be obtained in the following manner. According to the brand selling information in the brand information of each brand, the commodity selling price and the selling quantity of the commodities under each brand flag are accumulated to obtain the sum of the selling price of the commodities under each brand flag and the total selling quantity of the commodities, and the average price of the commodities sold under the brand flags is calculated by dividing the sum of the selling price of the commodities under the brand flags by the total selling quantity of the commodities under the brand flags. For example, for brand P1, the selling price and the selling quantity of the brand P1 under flag are accumulated according to the data in table 2, the sum of the selling prices of the brand P1 under flag is 7500 yuan, the total selling quantity of the brand P1 under flag is 200, the average selling price of the brand P1 under brand is 7500/200-37.5 yuan, which is expressed as "P1: 37.5".

According to this embodiment, the brand sold maximum bid amount and the brand sold minimum bid amount can be obtained in the following manner. The commodity names of commodities with highest commodity price and lowest commodity price are obtained from the brand commodity information of each brand, the commodity selling quantity of the commodities is obtained from the brand selling information according to the commodity names, the commodity selling quantity of the commodities with the highest commodity price is used as the brand selling price highest commodity quantity, and the commodity selling quantity of the commodities with the lowest commodity price is used as the brand selling price lowest commodity quantity. Taking brand P1 as an example, as shown in table 2, if the commodity price of commodity N50 is the highest and 120 yuan, the corresponding selling quantity of commodities is 0, the commodity price of commodity N1 is the lowest and 20 yuan, and the corresponding selling quantity of commodities is 2, then the selling quantity of the highest bid price commodity of brand P1 is 0, and the selling quantity of the lowest bid price commodity of brand 2 are represented as "P1: 0, 2" in combination.

In order to further improve the training efficiency of the deep learning neural network, the input data of the network is usually discretized in advance, and according to another embodiment of the present invention, after steps S110 and S120 are completed, the user feature information, the first brand feature information, and the second brand feature information are encoded. According to this embodiment, the One-Hot encoding can be performed on the information on the time length, the information on the price, the information on the geographical position, and the like in the above feature information. One-Hot encoding is a way to process data using an X-bit state register to encode X states, each with its own independent register bit and only One of which is active at any time. For example, the single brand dwell time is divided into segments according to 5 minutes, the segments are divided into 10 segments in total, each segment is initialized to 0000000000, and the number of the single brand dwell time falling in a certain segment is set to 1 from 0. At this time, if the residence time of a single brand is less than 5 minutes, the corresponding code is 1000000000, and if the residence time of a single brand is 5-10 minutes, the corresponding code is 0100000000.

Next, in step S230, a first deep learning neural network and a second deep learning neural network are constructed, respectively. According to one embodiment of the invention, a deep learning neural network is constructed through a TensorFlow framework, the number of layers of the deep learning neural network is set to be 5-9 layers according to the condition of input information, each layer is a convolutional layer, and specific parameters of each layer need to be adjusted in the training process. And after the calculation of each convolution layer is finished, an ELU function is used as an activation function, the last convolution layer is activated and then accessed to the full-connection layer, and the activation function is input to a Softmax function for processing, so that a first deep learning neural network and a second deep learning neural network are constructed. In the subsequent training process, the prediction result of the Softmax function and the real label value are used as input, a Cross-entropy algorithm is used for calculating a corresponding loss value, the loss value is input into a Momentum optimization method for calculating gradient, and model parameters of the deep learning neural network are updated. Example code for constructing the first deep-learning neural network and the second deep-learning neural network is as follows:

def conv(inputs,diameter,Nin,Nout,name):

fan_in＝diameter*diameter*Nin

print"WARNING:USING DIFFERENT STDDEV FOR CONV！"

stddev＝math.sqrt(1.0/fan_in)

kernel＝tf.Variable(tf.truncated_normal([diameter,diameter,Nin,Nout],

stddev＝stddev),name＝name+'_kernel')

return tf.nn.conv2d(inputs,kernel,[1,1,1,1],padding＝'SAME')

def inference(self,feature_planes,N,Nfeat):

NK＝192

NKfirst＝192

conv1＝ELU_conv_pos_dep_bias(feature_planes,5,Nfeat,NKfirst,N,

'conv1')

conv2＝ELU_conv_pos_dep_bias(conv1,3,NKfirst,NK,N,'conv2')

conv3＝ELU_conv_pos_dep_bias(conv2,3,NK,NK,N,'conv3')

conv4＝ELU_conv_pos_dep_bias(conv3,3,NK,NK,N,'conv4')

conv5＝ELU_conv_pos_dep_bias(conv4,3,NK,NK,N,'conv5')

conv6＝ELU_conv_pos_dep_bias(conv5,3,NK,NK,N,'conv6')

conv7＝ELU_conv_pos_dep_bias(conv6,3,NK,NK,N,'conv7')

conv8＝ELU_conv_pos_dep_bias(conv7,3,NK,NK,N,'conv8')

conv9＝ELU_conv_pos_dep_bias(conv8,3,NK,NK,N,'conv9')

after the first deep learning neural network and the second deep learning neural network are built, step S240 is performed, the user characteristic information, the first brand characteristic information and the second brand characteristic information are used as input of the first deep learning neural network, the user purchasing behavior is used as a label for supervised learning, and the first deep learning neural network is trained to obtain the user purchasing probability of each brand. According to one embodiment of the invention, for the first deep learning network based on the TensorFlow framework, the model input parameters comprise: filter size (1,2,4), number of filters (64,128,192), training step size (0.001-0.01), number of training iterations (60-1000 ten thousand), size of batch (16,64,128, 256). According to this embodiment, after the first deep learning neural network is trained, the probability that the user purchases brands P1-P100 is obtained.

Finally, in step S250, the user feature information, the second brand feature information, and the user purchase probability are used as inputs of a second deep learning neural network, and a walking route in a shopping mall is used as a label for supervised learning, and the second deep learning neural network is trained to predict a brand combination with the maximum user purchase probability. According to one embodiment of the invention, for the second deep learning network based on the TensorFlow framework, the model input parameters comprise: filter size (1,2,4), number of filters (64,128,192), training step size (0.001-0.01), number of training iterations (60-1000 ten thousand), size of batch (16,64,128, 256). According to this embodiment, after the second deep learning neural network is trained, the combination of cards that predict the highest probability of user purchase is P5, P20, and P77, which are denoted as [ P5, P20, P77 ]. Regarding the construction and training method of the deep learning neural network, the implementation process based on the tensrflow framework can be performed by the existing mature technical method, which is not described herein again.

If the training result of the first deep learning neural network or the second deep learning neural network is not ideal enough, weighting processing can be performed on input information in a mode of standardizing data through specified weight, but the weight needs to be controlled well when the training result is used, so that the influence of the activity, season and special conditions of malls or brands on user behaviors is fully considered, and the data are prevented from being damaged. According to another embodiment of the present invention, the user characteristic information and the first brand characteristic information are weighted according to a specific event, respectively. The specific events can be understood as events affecting user behaviors such as season changes, related activities performed by malls or brands during holidays, and when such specific events occur, certain positive causes including but not limited to seasons, holidays and the like should be reserved or amplified for the user characteristic information and the first brand characteristic information, and certain negative causes including but not limited to special cases such as epidemics and the like should be removed or reduced. Specifically, if a brand is doing activities, the total residence time of the brand, the average residence time of the brand, the purchasing behavior of the user, the average purchasing price of the goods, etc. may generate peaks, and in order to smooth the data, a weight smaller than 1 and larger than 0 is multiplied in front of the data to be used as a penalty. Conversely, if a special case occurs, the above data may have a low valley, and a weight greater than 1 needs to be multiplied in order to smooth the data.

According to another embodiment of the invention, the geographical position information of the user stored in the data storage device is generated by processing positioning data acquired by the prediction server based on Wi-Fi probe devices installed on each layer of the shopping mall, wherein the positioning data is the geographical position information of the mobile terminal of each mobile terminal detected by the Wi-Fi probe devices within a preset range. The prediction server is connected with the geographical position coordinates of the shopping malls to form shopping mall geographical fences after acquiring the geographical position information of the mobile terminals, the geographical position information of the mobile terminals is filtered through the shopping mall geographical fences, and the geographical position information of the mobile terminals obtained after filtering is used as the geographical position information of users of the users with the mobile terminals in the shopping malls.

Fig. 2 shows a schematic diagram of a prediction server 200 according to one embodiment of the invention. As shown in FIG. 2, prediction server 200 includes a data store 210, a first processing module 220, a second processing module 230, a building module 240, a first training module 250, and a second training module 260.

The data storage device 210 stores user geographical location information of each user holding the mobile terminal in a shopping mall, including user identification, longitude and latitude, a timestamp and a floor, brand information including a brand name, brand geographical location information, brand commodity information and brand selling information, and brand information, and basic demographic data. The brand commodity information comprises a brand flag commodity name, a commodity price, a commodity quantity, commodity shelf time, expected commodity shelf time and season response, the brand selling information comprises a brand flag commodity name, a commodity selling price, a commodity selling quantity and a user identification, and the basic demographic data comprises age, gender, income level and the industry to which the brand commodity information belongs.

The first processing module 220 is connected to the data storage device 210, and is adapted to obtain the user geographic location information and the brand information from the data storage device 210, and process the user geographic location information and the brand information to obtain user characteristic information and first brand characteristic information, where the user characteristic information includes a user purchasing behavior and a walking route in a shopping mall. The first processing module 220 is further adapted to process the brand name, the brand geographical location information and the user geographical location information to obtain a total residence time of the brand, a residence time of a single brand and a walking route in the mall; processing brand selling information, and acquiring user characteristic information by combining the residence time of a single brand and a walking route in a mall; and processing the total residence time of the brand and the walking route in the market to acquire the characteristic information of the first brand. The user characteristic information further comprises the average price of commodities under the brand flag, the stay time of a single brand and a frequent item set of purchased brands, the first brand characteristic information comprises the average stay time of the brands, the total stay time of the brands and an optimal market walking route, the optimal market walking route is the first K market walking routes with the highest repetition rate, and K is an integer not less than 1.

The second processing module 230 is coupled to the data storage device 210 and is adapted to obtain basic demographic data from the data storage device 210 and process the brand information, the user characteristic information, and the basic demographic data to obtain second brand characteristic information. The second brand feature information comprises the average price of the commodities under the brand flag, the brand popularity, the brand target user information, the brand geographical location information, the average price of the commodities sold by the brand, the highest quantity of the commodities sold by the brand and the lowest quantity of the commodities sold by the brand and marked with the price.

The building module 240 is adapted to build a first deep learning neural network and a second deep learning neural network, respectively.

The first training module 250 is respectively connected to the first processing module 220, the second processing module 230 and the building module 240, and is adapted to train the first deep learning neural network by taking the user feature information, the first brand feature information and the second brand feature information acquired from the first processing module 220 as inputs of the first deep learning neural network built by the building module 240 and taking the user purchasing behavior acquired from the first processing module 220 as a label for supervised learning, so as to acquire user purchasing probabilities of the brands.

The second training module 260 is connected to the first processing module 220, the second processing module 230, the building module 240 and the first training module 250, and is adapted to train the second deep learning neural network by using the user feature information acquired from the first processing module 220, the second brand feature information acquired from the second processing module 230 and the user purchase probability acquired from the first training module 250 as inputs of the second deep learning neural network built by the building module 240, and using the intra-shop walking route acquired from the first processing module 220 as a label for supervised learning, so as to predict a brand combination with the maximum user purchase probability.

Fig. 3 shows a schematic diagram of a prediction server 300 according to yet another embodiment of the invention. As shown in fig. 3, the data storage device 310, the first processing module 320, the second processing module 330, the building module 340, the first training module 350, and the second training module 360 of the prediction server 300 correspond to the data storage device 210, the first processing module 220, the second processing module 230, the building module 240, the first training module 250, and the second training module 260 of the prediction server 200 in fig. 2 one to one, and are consistent, and a weighting module 370 is added. The weighting module 370 is connected to the first processing module 320, the first training module 350, and the second training module 360, respectively, and is adapted to perform weighting processing on the user characteristic information and the first brand characteristic information acquired from the first processing module 350, respectively, according to a specific event.

Fig. 4 shows a schematic diagram of a prediction server 400 according to yet another embodiment of the invention. As shown in fig. 4, the data storage device 410, the first processing module 420, the second processing module 430, the building module 440, the first training module 450, and the second training module 460 of the prediction server 400 correspond to the data storage device 210, the first processing module 220, the second processing module 230, the building module 240, the first training module 250, and the second training module 260 of the prediction server 200 in fig. 2 one to one, are identical, and are added with an encoding module 480. The encoding module 480 is connected to the first processing module 420, the second processing module 430, the first training module 450, and the second training module 460, respectively, and is adapted to encode the user characteristic information, the first brand characteristic information, and the second brand characteristic information acquired from the first processing module 420 and the second processing module 430.

Fig. 5 shows a schematic diagram of a prediction server 500 according to yet another embodiment of the invention. As shown in fig. 5, the data storage device 510, the first processing module 520, the second processing module 530, the building module 540, the first training module 550, and the second training module 560 of the prediction server 500 correspond to the data storage device 210, the first processing module 220, the second processing module 230, the building module 240, the first training module 250, and the second training module 260 of the prediction server 200 in fig. 2 one to one, and are consistent, and a filtering module 590 is added. The filtering module 590 is connected to the data storage device 510, and is adapted to obtain the geographic location information of the mobile terminal of each mobile terminal, connect to the geographic location coordinates of the shopping mall to form a shopping mall geographic fence, filter the geographic location information of the mobile terminal through the shopping mall geographic fence, use the filtered geographic location information of the mobile terminal as the geographic location information of each user holding the mobile terminal in the shopping mall, and store the geographic location information of the user in the data storage device 510.

The specific steps and embodiments of the market brand combination prediction are disclosed in detail in the description based on fig. 1, and are not described herein again.

As for the existing market brand combination prediction method, the perfect combination of a large amount of market indoor positioning data accumulated on line and market online sales activities is not realized, and accurate and effective analysis and prediction processing on market customer group characteristics and market brand combinations are difficult to realize. According to the technical scheme of market brand combination prediction, user geographic position information and brand information are processed to obtain user characteristic information and first brand characteristic information, then the brand information, the user characteristic information and basic demographic data are processed to obtain second brand characteristic information, then the user characteristic information, the first brand characteristic information and the second brand characteristic information are input into a first deep learning network to be trained to obtain user purchase probability of each brand, and finally the user characteristic information, the second brand characteristic information and the user purchase probability are input into a second deep learning network to be trained to predict a brand combination with the maximum user purchase probability. In the technical scheme, the user geographic position information, the brand information and the basic demographic data are processed to obtain the relevant characteristic information of the user and the brand, the characteristic information is further processed through weighting processing, coding processing and the like to improve the effectiveness of the characteristic information, the constructed first deep learning network is utilized to analyze the characteristic information of the user so as to obtain accurate user images of a shopping mall customer group, and finally, a brand and market combination is predicted through the constructed second deep learning network, so that the market is guided to plan a more efficient and reasonable brand combination to meet the user requirements, further more accurate oriented marketing activities are provided, and the perfect combination of a large amount of indoor positioning data accumulated on line and offline marketing activities of the shopping mall is realized.

A9. The method of any one of a1-8, further comprising:

and respectively carrying out weighting processing on the user characteristic information and the first brand characteristic information according to a specific event.

A10. The method of any one of a1-9, further comprising:

and coding the user characteristic information, the first brand characteristic information and the second brand characteristic information.

A11. The method of any one of a1-10, further comprising:

acquiring geographical position information of the mobile terminals of each mobile terminal;

connecting the market geographic position coordinates to form a market geofence;

and filtering the geographical position information of the mobile terminal through the mall geo-fence, and taking the geographical position information of the mobile terminal obtained after filtering as the geographical position information of each user holding the mobile terminal in the mall.

B13. The forecast server according to B12, wherein the brand commodity information includes brand flag under commodity name, commodity price, commodity quantity, commodity shelf time, commodity expected shelf time and season response.

B14. The forecast server according to B12 or 13, wherein the brand selling information includes brand name under brand flag, selling price of the goods, selling quantity of the goods and user identification.

B15. The prediction server of any of B12-14, the basic demographic data comprising age, gender, income level, and industry of the genus.

B16. The prediction server of any one of B12-15, the first processing module further adapted to:

processing the brand name, the brand geographical position information and the user geographical position information to obtain the total residence time of the brand, the residence time of a single brand and a walking route in a mall;

processing the brand selling information, and acquiring the user characteristic information by combining the stay time of the single brand and the walking route in the mall;

and processing the total residence time of the brand and the walking route in the market to obtain the characteristic information of the first brand.

B17. The prediction server of any of B12-16, the user characteristic information further comprising: average purchase price of the goods, individual brand dwell time, and a frequent set of purchased brands.

B18. The prediction server of any one of B12-17, wherein the first brand feature information includes an average brand staying time, a total brand staying time ratio, and an optimal mall walking route, where the optimal mall walking route is the first K highest in repetition rate in the mall walking route, and K is an integer not less than 1.

B19. The prediction server of any one of B12-18, the second brand feature information comprising: the method comprises the following steps of average commodity price under a brand flag, brand popularity, brand target user information, brand geographical location information, average commodity price sold by the brand, the highest commodity quantity sold by the brand and the lowest commodity quantity sold by the brand.

B20. The prediction server of any of B12-19, further comprising a weighting module adapted to:

B21. The prediction server of any of B12-20, further comprising an encoding module adapted to:

B22. The prediction server of any of B12-21, further comprising a filtering module adapted to:

In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art will appreciate that the modules or units or components of the devices in the examples disclosed herein may be arranged in a device as described in this embodiment or alternatively may be located in one or more devices different from the devices in this example. The modules in the foregoing examples may be combined into one module or may be further divided into multiple sub-modules.

Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.

Furthermore, some of the described embodiments are described herein as a method or combination of method elements that can be performed by a processor of a computer system or by other means of performing the described functions. A processor having the necessary instructions for carrying out the method or method elements thus forms a means for carrying out the method or method elements. Further, the elements of the apparatus embodiments described herein are examples of the following apparatus: the apparatus is used to implement the functions performed by the elements for the purpose of carrying out the invention.

The various techniques described herein may be implemented in connection with hardware or software or, alternatively, with a combination of both. Thus, the methods and apparatus of the present invention, or certain aspects or portions thereof, may take the form of program code (i.e., instructions) embodied in tangible media, such as floppy diskettes, CD-ROMs, hard drives, or any other machine-readable storage medium, wherein, when the program is loaded into and executed by a machine, such as a computer, the machine becomes an apparatus for practicing the invention.

In the case of program code execution on programmable computers, the computing device will generally include a processor, a storage medium readable by the processor (including volatile and non-volatile memory and/or storage elements), at least one input device, and at least one output device. Wherein the memory is configured to store program code; the processor is configured to execute the store brand combination prediction method of the present invention according to instructions in the program code stored in the memory.

By way of example, and not limitation, computer readable media may comprise computer storage media and communication media. Computer-readable media includes both computer storage media and communication media. Computer storage media store information such as computer readable instructions, data structures, program modules or other data. Communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. Combinations of any of the above are also included within the scope of computer readable media.

As used herein, unless otherwise specified the use of the ordinal adjectives "first", "second", "third", etc., to describe a common object, merely indicate that different instances of like objects are being referred to, and are not intended to imply that the objects so described must be in a given sequence, either temporally, spatially, in ranking, or in any other manner.

While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this description, will appreciate that other embodiments can be devised which do not depart from the scope of the invention as described herein. Furthermore, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the appended claims. The present invention has been disclosed in an illustrative rather than a restrictive sense, and the scope of the present invention is defined by the appended claims.

Claims

1. A shopping mall brand combination prediction method adapted to be executed in a prediction server including a data storage device in which user geographical location information of each user holding a mobile terminal in a shopping mall including a user identification, a latitude and longitude, a timestamp, and a floor, brand information including a brand name, brand geographical location information, brand commodity information, and brand selling information are stored, brand information, and basic demographic data, the method comprising:

acquiring the user geographical position information and the brand information from the data storage device, processing the brand name, the brand geographical position information and the user geographical position information, acquiring the total residence time of the brand, the residence time of a single brand and the walking route in a shopping mall, processing the brand selling information, and acquiring user characteristic information by combining the residence time of the single brand and the walking route in the shopping mall;

processing the total residence time of the brand and the walking route in the market to obtain first brand characteristic information, wherein the first brand characteristic information comprises average residence time of the brand, the total residence time of the brand and an optimal market walking route, the optimal market walking route is the first K places with the highest repetition rate in the market walking route, K is an integer not less than 1, the user characteristic information comprises user purchasing behaviors of users, the walking route in the market, average commodity purchasing prices, residence time of a single brand and a frequent item set of purchased brands, and the walking route in the market is a brand name set for each user staying one by one in one day;

obtaining basic demographic data from the data storage device, and processing the brand information, the user characteristic information and the basic demographic data to obtain second brand characteristic information, wherein the second brand characteristic information comprises: the method comprises the following steps of (1) average commodity price under a brand flag, brand popularity, brand target user information, brand geographical location information, average commodity price sold by the brand, highest commodity quantity sold by the brand and lowest commodity quantity sold by the brand;

respectively constructing a first deep learning neural network and a second deep learning neural network;

taking the user characteristic information, the first brand characteristic information and the second brand characteristic information as input of the first deep learning neural network, taking the user purchasing behavior as a label for supervised learning, and training the first deep learning neural network to obtain the user purchasing probability of each brand;

and taking the user characteristic information, the second brand characteristic information and the user purchase probability as the input of the second deep learning neural network, taking the walking route in the mall as a label for supervised learning, and training the second deep learning neural network to predict the brand combination with the maximum user purchase probability.

2. The method of claim 1, wherein the branded goods information comprises a brand flag goods name, a goods price, a goods quantity, a goods listing time, a goods expected listing time, and whether season is due.

3. The method of claim 1, the brand selling information comprising a brand flag commodity name, a commodity sale price, a commodity sale quantity, and a user identification.

4. The method of claim 1, wherein the basic demographic data includes age, gender, income level, and industry of the subject.

5. The method of claim 1, further comprising:

6. The method of claim 5, further comprising:

7. The method of any of claims 1-6, further comprising:

8. A prediction server, the prediction server comprising:

the system comprises a data storage device, a server and a terminal, wherein the data storage device stores user geographic position information, brand information and basic demographic data of each brand of each user holding the mobile terminal in a shopping mall, the user geographic position information comprises user identification, longitude and latitude, timestamps and floors, and the brand information comprises brand names, brand geographic position information, brand commodity information and brand selling information;

the first processing module is suitable for acquiring the user geographic position information and the brand information from the data storage device, processing the brand name, the brand geographic position information and the user geographic position information, acquiring a total brand staying time, a single brand staying time and a walking route in a market, processing the brand selling information, acquiring user characteristic information by combining the single brand staying time and the walking route in the market, and processing the total brand staying time and the walking route in the market to acquire first brand characteristic information, wherein the first brand characteristic information comprises an average brand staying time, a total brand staying time ratio and an optimal market walking route, the optimal market walking route is the first K with the highest repetition rate in the market walking route, K is an integer not less than 1, and the user characteristic information comprises a user purchasing behavior, Travel routes in the store, average purchase price of goods, dwell time of individual brands, and frequent itemsets of purchased brands;

a second processing module adapted to obtain basic demographic data from the data storage device, and process the brand information, the user characteristic information, and the basic demographic data to obtain second brand characteristic information, wherein the second brand characteristic information includes: the method comprises the following steps of (1) average commodity price under a brand flag, brand popularity, brand target user information, brand geographical location information, average commodity price sold by the brand, highest commodity quantity sold by the brand and lowest commodity quantity sold by the brand;

the building module is suitable for respectively building a first deep learning neural network and a second deep learning neural network;

the first training module is suitable for taking the user characteristic information, the first brand characteristic information and the second brand characteristic information as input of the first deep learning neural network and taking the user purchasing behavior as a label for supervised learning, and training the first deep learning neural network to obtain the user purchasing probability of each brand;

and the second training module is suitable for training the second deep learning neural network by taking the user characteristic information, the second brand characteristic information and the user purchase probability as the input of the second deep learning neural network and taking the walking route in the mall as a label for supervised learning so as to predict the brand combination with the maximum user purchase probability.

9. The prediction server of claim 8, the branded goods information comprising brand flag under goods name, goods price, goods quantity, goods on shelf time, goods expected off shelf time, and season response.

10. The prediction server of claim 8, the brand sale information comprising a brand under-flag item name, an item sale price, an item sale quantity, and a user identification.

11. The prediction server of claim 8, the basic demographic data including age, gender, income level, and industry of the home.

12. The prediction server of claim 8, further comprising a weighting module adapted to:

13. The prediction server of claim 12, further comprising an encoding module adapted to:

14. The prediction server of any one of claims 8-13, further comprising a filtering module adapted to:

15. A computing device, comprising:

at least one processor; and

at least one memory including computer program instructions;

the at least one memory and the computer program instructions are configured to, with the at least one processor, cause the computing device to perform the method of any of claims 1-7.