JP2016207020A - Marketing method and system based on information entropy of behavior data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a marketing method and system in which the behavior data concerning the purchasing trend of users is quantitatively calculated as entropy and the time change is measured, and thereby the long-term change of the users becoming targets is captured to estimate a current state.SOLUTION: A marketing method is characterized by estimating a purchasing trend of commodities or services of the present invention based on information entropy of the behavior data of purchasers. The marketing method includes: a step of generating an information transition probability matrix concerning the commodities or services as probability to purchase other commodities or services; a step of calculating the entropy (average information content) as the uncertainty of the transition probability matrix; a step of determining the change of the entropy by calculating the entropy again for the same commodity or service when arriving at a predetermined time; and a step of extracting new commodities or services from a database based on the determination of the change of the entropy.SELECTED DRAWING: Figure 8

Description

  The present invention relates to a marketing method and system based on information entropy of behavior data.

  In the conventional method of marketing purchase trends of goods or services based on user behavior data, segmentation is based on the latest user's hobbies and preferences, classifying user attributes by estimation, and predicting in the future by patterned behavior analysis, Marketing measures such as recommendations and campaigns are implemented for each classified segmentation. However, with this method, user attributes, especially hobbies and preferences, always change, and even if simply estimated by the latest behavior data, the correct result Cannot be obtained, and the result largely fluctuates due to superiority or inferiority of the estimation logic, and an effective result is not necessarily obtained.

  Patent Document 1 discloses a configuration in which a user has a prior probability of purchasing a product, an estimated probability that a previously purchased product is a specific product, and a weight is estimated by combining them using the maximum entropy principle to recommend the product. However, in this case too, it does not grasp the long-term personal purchasing trend.

JP 2008-287550 A

  The present invention has been made in view of the above-described conventional problems, and an object of the present invention is to obtain behavioral data relating to a user's purchase trend in order to quantitatively grasp a change in the user's long-term purchase trend. It is to provide a marketing method and system that can calculate information entropy, estimate the current state of interest by measuring its temporal change, and implement measures suitable for each state. .

  A marketing method according to one aspect of the present invention for achieving the above object is a marketing method for estimating a purchase trend of a product or service based on information entropy of purchaser's behavior data. A step of generating an information transition probability matrix related to a product or service as a probability of purchase, a step of calculating entropy (average amount of information) as an uncertainty of the transition probability matrix, and the same product or A step of calculating entropy again for the service and determining a change in entropy; and a step of extracting a new product or service from a preset database based on the determination of the change in entropy.

The entropy change may include entropy increase, status quo, and decrease.
The information on the product or service includes the number of visits at the store or website where the product or service is purchased, the visit frequency, the staying time, the number of new visitors, the number of revisitors, the degree of recognition, the number of inflow pages, the access log, the number of sessions, One or more of cookie information, search terms, ad recency, ad frequency, website conversions or rates, SNS (social networking service) engagements or rates, and SNS viral numbers or rates May be included.

  In order to achieve the above object, a marketing system according to one aspect of the present invention is a marketing system including at least a database that estimates purchase trends of goods or services based on information entropy of purchaser behavior data, and includes a communication network. A Web server that collects information related to the purchase of goods or services from a plurality of websites connected to the website, and a transition probability matrix generation means that generates an information transition probability matrix related to the goods or services as the probability of purchasing another product or service; Entropy calculating means for obtaining entropy (average amount of information) as uncertainty of the transition probability matrix, and determining entropy change by calculating entropy again for the same product or service when a predetermined time is reached An entropy change judging means; New products or services based on the determination of the change in serial entropy and a product extraction means for extracting from the database.

  The marketing system personalizes a new product or service extracted by the product extraction means for each browsing user of the Web site, and cooperates with a marketing action tool including either advertisement distribution, recommendation, or marketing mail distribution. A marketing cooperation means is further provided.

  According to the present invention, in order to quantitatively grasp the long-term change of the user, the information entropy of the action data of the specific user is calculated using big data, and the temporal change is measured. The current state of the user is estimated, and by implementing measures suitable for each of the information entropy increase, current state maintenance, and decrease states, the measure can be made more efficient than before.

It is a conceptual diagram for demonstrating the marketing method by this invention. It is a conceptual diagram by an example for demonstrating information entropy. It is a conceptual diagram by an example for demonstrating the change of information entropy. It is a figure for demonstrating the example of calculation of information entropy. It is a schematic structure figure showing a marketing system by one embodiment of the present invention. It is a schematic block diagram of the calculation and analysis server shown in FIG. It is a schematic block diagram which shows the information entropy calculation system by one Embodiment of this invention. 3 is a flowchart illustrating a marketing method according to an exemplary embodiment of the present invention.

  Hereinafter, specific examples of embodiments for carrying out the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a conceptual diagram for explaining a marketing method according to the present invention.

  Referring to FIG. 1, for example, a single male 20-year-old who has an annual income of 3 million is SNS (social network), etc. Participate in, buy a car (foreign car) with a loan, drive domestically with friends you met through SNS, etc., and also travel abroad relying on overseas acquaintances. After that, he married in his 30s, rented a house for a couple at the age of 35 and traveled with the couple. A man in his 40s considers future savings by replacing cars with domestic cars. After that, he earns 9 million a year, and makes a loan for childcare, purchases land and a detached house, and joins life insurance.

  In this way, the user changes while receiving various influences in his life, and the purchasing trend of products and services accompanying this changes. The federated marketing service is a marketing service that uses multiple cloud services such as SNS (social networking service) and online sales in an integrated manner. Each of the products and services includes not only purchase history but also search history. Leverage your cloud customer profile values. Marketing activities based on customer profiles are customer behavior-driven based on user trends.

  The present invention captures a long-term change in profits that a customer brings to a company over a predetermined period, that is, LTV (customer lifetime value), and holds communication activities of users including 3rd parties as behavior data for a long period of time, By capturing the signs, campaigns that were not possible in the past are conducted and used for marketing based on a long-term perspective such as marriage, home, and car. Note that user data over a long period of time uses big data, and is combined using some key that can identify the user such as a member ID in the same site. About the method of specifying a user from big data, since it remove | deviates from the summary of invention, the specific structure and description are abbreviate | omitted.

  People's hobbies and preferences change like a cycle over a certain period of time, and even now they may not be interested but may be interested for some reason. Predict and make recommendations. For example, users with similar “hobbies and preferences” are segmented from long-span behavior data, and what kind of behavior will be predicted in the future is predicted. As an example, see the home site as a standard type → see the home appliance site → see the car site → see the baby goods related site, see the travel site as a hobby pursuit type → see the camera related site → apparel For example, see related sites.

  Entropy, on the other hand, can be set as a measure of how difficult a certain event will occur. In situations where consumer purchase behavior can occur on a regular basis, such as regular purchases, entropy decreases and the amount of information is stable. It can be said that it is in a state. When entropy rises due to factors such as irregular or sudden purchase behavior due to multiple dimensions, it is assumed that the amount of information is chaotic due to its characteristics. The present invention intends to digitize this large amount of chaotic information using a calculation formula based on the level of entropy and utilize it for marketing.

  FIG. 2 is a conceptual diagram illustrating an example of information entropy.

  Referring to FIG. 2, the information entropy will be described by exemplifying a case where the user purchases drinking water. The user selects carbonated drinking water from tea drinking water, mineral water, and carbonated drinking water, and carbonated drinking water. Assuming that Cola A was always purchased monthly from colas, ciders, and beers in the market, information entropy is a factor that captures signs of attitude change that the user intends to purchase Cider B at some chance Utilize.

  In the conventional case where marketing measures are implemented for each segmentation classified by estimation of individual attributes, the results vary greatly depending on the superiority or inferiority of the estimation logic. Measure the change and estimate the current state of the target user, and implement measures appropriate to the information entropy “rising”, “maintaining”, and “falling” Increase the success rate of marketing measures more efficiently than before.

  Select one of the specific patterns belonging to the user group that segmented the users who are similar to “changes in hobbies and preferences” from a plurality of specific patterns such as the standard type or the hobby pursuit type, An information transition probability matrix is generated as a probability of purchasing another product or service from a plurality of corresponding products or services stored in a database.

  Next, an average information entropy is calculated as an index representing the uncertainty of the generated information transition probability matrix. By calculating the logarithm of the reciprocal of establishing the amount of information when an event of occurrence probability (occurrence probability) occurs, the self-information amount of the event is defined, and the average information entropy is calculated by calculating the average value for all events Is done. This means that an average value (expected value) of a set having stochastic events as elements is obtained by multiplying the self-information amount of each event, that is, the value of each random variable by the occurrence probability. In information theory, entropy is defined as the number of bits per symbol by setting the base of the logarithm to 2 so that the consistency with computer computation is high. In the present specification, this is called information entropy. In the present invention, this information entropy is calculated for each predetermined time, and the change in information entropy is determined based on time-series data that changes with time.

  FIG. 3 is a conceptual diagram illustrating an example of the change in information entropy.

  Referring to FIG. 3, information entropy of individual users is calculated and held for each month, and a change in information entropy of a group to which the user belongs is determined by comparing calculation results of specific months. The upper left side shows the number of purchases of each product A to D by each user A to C in N month, the lower left side shows the number of purchases of each product A to D by each user A to C in N-N month, and each user A change in information entropy obtained by averaging the information entropies of a plurality of users of the user group to which the user belongs is shown on the right side. In the state determination, an upward arrow “↑” indicates an increase in information entropy, a right arrow “→” indicates maintenance of information entropy, and a downward arrow “↓” indicates a decrease in information entropy. Based on this judgment, implement appropriate marketing measures. For example, when the information entropy is rising, the information entropy is in an active state, so it is determined that there is a high possibility of accepting a new proposal that matches the user's hobby preference.

  The behavioral data used to calculate information entropy includes, for example, the number of visits to the store or website where the product or service is purchased, the visit frequency, the staying time, the number of new visitors, the number of revisiters, the degree of recognition, the number of inflow pages, and access Use various information such as logs, number of sessions, cookie information, search terms, ad recency, ad frequency, conversions or rates on websites, engagements or rates on SNS, and viral numbers or rates on SNS Is possible.

  FIG. 4 is a diagram for explaining a calculation example of information entropy.

  FIG. 4 shows an example of purchasing beer as an example of online purchase. The attribute information of the user A is, for example, a woman in her 30s when collecting data from an access log of an online sales site, and the number of purchases per month is five. As an example of an individual user, a change in attitude is examined from the purchase history of user A for three months. Which brand of beer is purchased (share) is determined from the number of purchases for each company in October, November, and December, an information transition probability matrix is generated, and information entropy is calculated based on the information amount. The information entropy in October is as high as 2.0, stable in November in 0.0, and high in 1.7 in December.

  The calculation of information entropy is performed for a plurality of users. If the information entropy of user B is calculated (not shown) in the same manner as user A, for example, the information entropy shows the minimum value, so it is decided which one to buy. In the case of user C, who seems to be a user and is highly deterministic, and that has been purchased evenly by any manufacturer (not shown), it seems that he has not decided which one to buy because the information entropy shows the maximum value It is determined that the user is highly uncertain. In this way, even when there are many options, it is possible to mechanically determine the user's trend based only on the information entropy regardless of the purchase amount.

  As prediction based on long span action data, a change in information entropy is determined from the result of monthly information entropy. In the example of FIG. 4, the purchase results of beer by brand are quantified and the future psychological state of purchase is estimated. For example, as a result of statistically investigating the group to which the user A belongs, it is found that there are many users who are easily affected by word of mouth as an attribute of information entropy. When a change in information entropy is determined in the same manner as for user A, for example, user B obtains information entropy for each visitor of the online sales site, for example, information entropy for 1000 people, as with user A, so that each user's information Entropy attributes are determined. By determining the attribute of information entropy, it becomes possible to manage the user and the user group by segmenting the attribute for each attribute, and use it for the measures related to the market for the individual users belonging to the user group.

  Although it is not possible to express the state of “I don't know which one to choose” simply by the number of visits of the access log of the website in the past, for example, when visiting the access log of the website into a category and visiting only a specific category When it is found that the information entropy is decreasing and the information entropy increases, it is determined that the interest of the category is in the MAX state, and the marketing measure is implemented. Or, when the category of the visited website changes, such as marriage → car → house, the purchase including the psychological state of the user based on the change of information entropy is not performed instead of simply predicting from the previous visited place as in the past. Trends are estimated, it is judged whether the visit is based on a clear intention rather than a whimsical visit, and effective measures can be implemented. In order to utilize for marketing, PDCA (plan (plan) -do (execution) -check (evaluation) -act (improvement) based on purchase trends of users belonging to a preset user group, for example, from user trends ) We will implement measures such as recommending products or services by passing information to marketers who have the ability to cycle, but here we will omit detailed explanations of marketing measures because they are out of the scope of the invention. .

  FIG. 5 is a schematic configuration diagram illustrating a marketing system according to an embodiment of the present invention. FIG. 6 is a schematic configuration diagram of the calculation / analysis server shown in FIG.

  5 and 6, the marketing system based on the information entropy of behavior data according to the present embodiment includes a plurality of websites 1, a web server 2, a database (DB) server 3, and a measurement / analysis server 4 in a communication network. 5 and connected.

  The Web site 1 is a Web site that is accessed by an individual user at the time of online purchase or the like, and is a plurality of Web sites that are accessed by a plurality of users including a plurality of Web sites accessed by the individual user.

  The web server 2 collects information such as access logs and cookies from a plurality of websites and associates and identifies the users. At this time, not only the website where the user directly purchased the product or service online, but also the information searched by the user, the SNS, etc. engagement and viral (number of shares, number of comments, word of mouth, spread rate, number of likes, etc.) Etc. are also used by analyzing their contents.

  The database server 3 stores user information collected via the Web server 2 or the like. Individual users are segmented for each user who has a similar “change in hobbies and preferences” from among a plurality of preset specific patterns, and are distributed to corresponding specific user groups.

  The calculation / analysis server 4 extracts information stored by segmenting users who have similar hobbies and preferences with respect to a specific user from a large amount of information stored in the DB server 3 and purchasing a product or service. The trend is quantitatively calculated by information entropy, the current state of the target user is estimated, and the change is measured over a long period of time. Implement marketing measures. The analysis results are reflected in various marketing actions such as advertisements for websites accessed by users, recommendations for online sales, and personalization by e-mail.

  Referring to FIG. 6, the calculation / analysis server 4 includes control means 41, storage means 42, input / output means 43, communication means 44, transition probability matrix generation means 45, entropy calculation means 46, entropy change determination means 47, product extraction. Means 48 and marketing cooperation means 49. The calculation / analysis server 4 is configured by a computer system, and implements each means (45 to 49) by mounting an application linked with hardware and software.

  The control means 41 controls various processes of the calculation / analysis server 4 by a CPU (central processing unit), extracts information from the database server 3, calculates user trends by information entropy, analyzes long-span action data, and implements marketing measures. Etc. are executed.

  The storage means 42 includes a storage device such as a ROM, a RAM, and an HDD (hard disk drive), and each means (45 to 49) that functions in the present embodiment uses related hardware by installing an application program. This is implemented by software processing.

  The input / output means 43 includes an input / output device (not shown) such as a touch panel that combines a pointing device such as a general keyboard and mouse and a monitor, or an input device such as a touch pad and a display device such as an LCD display. , Mainly used as an administrator or operator.

  The communication means 44 connects the calculation / analysis server 41 to a plurality of Web sites 1, Web servers 2, and database servers 3 via the communication network 5, and is used by each means (45 to 49). The communication network 5 is not limited to a wired connection such as an optical fiber or metal wiring, but includes a mobile phone communication network, an LTE communication network, a wireless LAN, and the like.

  The transition probability matrix generation means 45 refers to the database server 3 and, for example, “changes in hobbies and preferences” are similar to a specific user from among a plurality of specific patterns such as a standard type or a hobby pursuit type set in advance. A specific pattern of a user group obtained by segmenting existing users is selected, and an information transition probability matrix is generated as a probability of purchasing another product or service from a plurality of corresponding products or services stored in a database in advance. Behavior data for generating the information transition probability matrix includes the number of visits to the actual purchase store or online sales site of the product or service, the visit frequency, the staying time, the number of new visitors, the number of revisitors, the degree of recognition, and online sales. The number of inflowing pages on the site, access logs, number of sessions, cookie information, search terms, frequency of advertisements, frequency of advertisements, number of conversions or rates on websites, number of engagements or rates in SNS, number of viral numbers or rates in SNS, etc. it can.

  The entropy calculating means 46 calculates information entropy (average information amount) as an index representing the uncertainty of the information transition probability matrix generated by the transition probability matrix generating means 45. The information entropy calculated by calculating the average value (expected value) for all the self-information amount of the occurrence probability event was calculated by eliminating human judgment regardless of the number of options and the amount of purchase price. It becomes possible to judge a user trend only by a value.

  The entropy change determination unit 47 compares the information entropy calculated by the entropy calculation unit 46 at a predetermined time interval, for example, every month, and determines information entropy that changes with time based on time-series data. The determination of the change in information entropy is classified into, for example, “up”, “maintain current”, and “down”.

  The product extraction unit 48 extracts from the database 3 a new product or service suitable for each change state of “increased”, “maintenance of current”, and “down” of the information entropy determined by the entropy change determination unit 47. The determination of information entropy change is performed for many users, and the attribute of information entropy for each user is determined. By determining the attribute of information entropy, the user and its user group are managed by segmenting each attribute, and other products or services purchased by users belonging to the user group are extracted for individual users.

  The marketing cooperation means 49 distributes advertisements personalized for each browsing user of the Web site 1 through the communication means 44 for the new product or service extracted by the product extraction means 48, recommendations on the online sales site, electronic Provide information in cooperation with various marketing action tools such as personalization mail delivery by email.

  FIG. 7 is a schematic configuration diagram illustrating an information entropy calculation system according to an embodiment of the present invention.

  Referring to FIG. 7, the information entropy calculation system includes a transition probability matrix generation unit 45, an entropy calculation unit 46, an entropy change determination unit 47, which is realized by executing an application implemented in the calculation / analysis server shown in FIG. Product extraction means 48 and marketing cooperation means 49 are included.

  As described in FIG. 4, the information entropy calculation system collects monthly data (11a to 11n) of products or services purchased through online sales as action log data, for example, as a long span action history for each user. An information transition probability matrix is generated based on the entropy calculation parameter 12, and the purchase amount and the share are scored (13), and the result is stored in the database server 3. As described above, the entropy calculation parameter 12 includes the number of visits to the store or online sales site, visit frequency, stay time, number of new visitors, number of revisitors, recognition, number of inflow pages, access logs, number of sessions. Cookie information, search terms, advertisement recency, frequency of advertisements, website conversions, SNS engagements and virality.

  The action log data is scored (13) based on the generated information transition probability matrix, and the information entropy is calculated (14) based on the amount of information, and the monthly change, for example, “rise”, “maintenance” , “Descent” is determined. By measuring the change in information entropy, the current state of the target user is estimated, and measures suitable for each state of information entropy are implemented. The calculation of information entropy is performed for a plurality of users.

  The marketing tool (15) determines the entropy attribute from the change in entropy and refers to the purchase trend of the same user group segmented for each attribute stored in the database server 3, for example, to a preset user group Extract new products or services based on the purchase trends of the users to which they belong, and various kinds of marketing such as advertisement distribution personalized for each user by PDCA cycle, recommendations on online sales sites, email distribution by e-mail, etc. Implement marketing measures.

  FIG. 8 is a flowchart illustrating a marketing method according to an embodiment of the present invention.

  Referring to FIG. 8, a process flow from finding the regularity of hobbies and tastes by determining a change in information entropy from the above-described user long span behavior data, predicting the future, and recommending marketing measures is shown.

  The marketing method by the information entropy calculation system implemented in the calculation / analysis server 4 is based on the access logs of a plurality of websites 1 connected to the communication network 5 via the web server 2 or the like for the goods or services purchased by the user. Collect information related to the purchase of goods or services for each user using cookie information, etc., and refer to the database server 3, and users who have similar purchasing trends among a plurality of preset specific patterns One of the specific patterns belonging to the segmented user group is selected and one of the corresponding user groups is selected (step S101).

  The behavior data is extracted based on the specific pattern of the user group selected from the database server 3, the entropy calculation parameter 12 is specified, the behavior log data is scored from the long span behavior history of the user and stored in the database. An information transition probability matrix is generated as a probability of purchasing another product or service from the corresponding plurality of products or services (step S102).

  As an index representing the uncertainty of the generated information transition probability matrix, an average value (expected value) is obtained for all self-information amounts of occurrence probability events, and information entropy is calculated (step S103).

  The calculated information entropy is compared every month, and the information entropy that changes with time is determined based on time-series data, and is classified into “rising”, “maintaining current”, and “decreasing” (step S104).

  The attribute of the user's information entropy is determined according to the determined change state of the information entropy, and a new product with reference to the purchase trend of the same user group segmented for each attribute stored in the database server 3 Alternatively, services are extracted (step S105).

  The marketer personalizes the extracted new product or service for each user, for example, distribution of advertisements personalized for each browsing user of the website 1, recommendation on the online sales site, personalization distribution by e-mail, etc. Information is provided in cooperation with various marketing action tools (step S106).

  As mentioned above, although embodiment of this invention was described in detail, referring drawings, this invention is not limited to the above-mentioned embodiment, In the range which does not deviate from the technical scope of this invention, it changes variously. It is possible to implement.

DESCRIPTION OF SYMBOLS 1 Web site 2 Web server 3 Database (DB) server 4 Calculation / analysis server 5 Communication network 11a-11n Long span action history 12 Entropy calculation parameter 13 Action log data scoring 14 Entropy calculation 15 Marketing tool 41 Control means 42 Storage means 43 Input / output means 44 Communication means 45 Transition probability matrix generation means 46 Entropy calculation means 47 Entropy change determination means 48 Product extraction means 49 Marketing cooperation means

Claims (5)

A marketing method for estimating purchase trends of goods or services based on information entropy of purchaser behavior data,
Generating an information transition probability matrix for the product or service as a probability of purchasing another product or service;
Calculating entropy (average amount of information) as uncertainty of the transition probability matrix;
Calculating entropy again for the same product or service when a predetermined time is reached, and determining a change in entropy;
Extracting a new product or service from a preset database based on the determination of the change in entropy.   The marketing method according to claim 1, wherein the entropy change includes an entropy increase, a current state maintenance, and a decrease.   The information on the product or service includes the number of visits at the store or website where the product or service is purchased, the visit frequency, the staying time, the number of new visitors, the number of revisitors, the degree of recognition, the number of inflow pages, the access log, the number of sessions, One or more of cookie information, search terms, ad recency, ad frequency, website conversions or rates, social network services (SNS) engagements or rates, and SNS viral numbers or rates The marketing method according to claim 1, further comprising: A marketing system including at least a database for estimating a purchase trend of a product or service based on information entropy of purchaser behavior data,
A web server that collects information regarding purchase of goods or services from a plurality of websites connected to a communication network;
A transition probability generating means for generating an information transition probability matrix related to the product or service as a probability of purchasing another product or service;
Entropy calculating means for obtaining entropy (average amount of information) as uncertainty of the transition probability matrix;
Entropy change determination means for determining entropy change by calculating entropy again for the same goods or services when a predetermined time is reached;
A marketing system comprising: product extraction means for extracting a new product or service from the database based on the determination of the change in entropy.   Marketing cooperation means for personalizing a new product or service extracted by the product extraction means for each browsing user of the Web site and for linking with a marketing action tool including either advertisement distribution or recommendation or marketing mail distribution The marketing system according to claim 4, wherein the marketing system is provided.

Images