WO2020048062A1 - 产品销售的智能推荐方法、装置、计算机设备和存储介质 - Google Patents

产品销售的智能推荐方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2020048062A1
WO2020048062A1 PCT/CN2018/124392 CN2018124392W WO2020048062A1 WO 2020048062 A1 WO2020048062 A1 WO 2020048062A1 CN 2018124392 W CN2018124392 W CN 2018124392W WO 2020048062 A1 WO2020048062 A1 WO 2020048062A1
Authority
WO
WIPO (PCT)
Prior art keywords
customer
product
information
vector matrix
recommendation
Prior art date
Application number
PCT/CN2018/124392
Other languages
English (en)
French (fr)
Inventor
金戈
徐亮
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020048062A1 publication Critical patent/WO2020048062A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis

Definitions

  • the present application relates to the field of computers, and in particular, to a smart recommendation method, device, computer device, and storage medium for product sales.
  • the main purpose of this application is to provide a smart recommendation method, device, computer equipment, and storage medium that can effectively intelligently recommend product sales.
  • this application proposes a smart recommendation method for product sales, including:
  • the first product vector matrix Inputting the first product vector matrix into a reverse recommendation model trained based on the LSTM model to output a first representation layer vector corresponding to the first product vector matrix; the first representation layer vector is corresponding to a customer Representation layer vector of the first word vector matrix of the retrieved information;
  • the customer's information includes at least the customer's contact information
  • This application also provides an intelligent recommendation device for product sales, including:
  • a first vectorization unit configured to vectorize product information to obtain a first product vector matrix
  • An inverse learning unit configured to input the first product vector matrix into a reverse recommendation model trained based on the LSTM model to output a first representation layer vector corresponding to the first product vector matrix;
  • the first The presentation layer vector is a presentation layer vector of the first word vector matrix corresponding to the retrieval information of the customer;
  • a search unit configured to search a preset customer database for the first word vector matrix that has a similarity with the first representation layer vector that meets a specified requirement, wherein the first word vector matrix and the customer information Associatedly stored in the customer database, the customer's information includes at least the customer's contact information;
  • An extraction recommendation unit is configured to extract information of a customer corresponding to the first word vector matrix, and recommend the product information to the customer according to the extracted contact method of the customer.
  • the present application further provides a computer device including a memory and a processor, where the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, implements the steps of any one of the foregoing methods.
  • the present application further provides a computer non-volatile readable storage medium having computer readable instructions stored thereon, and the computer readable instructions, when executed by a processor, implement the steps of the method described in any one of the above.
  • the intelligent recommendation method, device, computer equipment and storage medium for product sales of the present application calculate a representation layer vector corresponding to the representation layer vector according to the vectorization matrix of the product information of the product, and then find the representation layer vector.
  • the first word vector matrix with similarity that meets the specified requirements, and the first word vector matrix is a vector matrix obtained by vectorizing the customer's search information, so the contact information of the corresponding customer can be found according to the first word vector matrix.
  • Information is the information that the customer actively enters, so the customer corresponding to the first word vector matrix found should have a relatively high purchase desire and corresponding purchasing ability, so send new products to the corresponding customers and recommend resale effectiveness. higher.
  • FIG. 1 is a schematic flowchart of a smart recommendation method for product sales according to an embodiment of the present application
  • FIG. 2 is a schematic block diagram of a structure of a smart recommendation device for product sales according to an embodiment of the present application
  • FIG. 3 is a schematic block diagram of a structure of a computer device according to an embodiment of the present application.
  • an embodiment of the present application first provides a smart recommendation method for product sales, including steps:
  • the preset customer database for the first word vector matrix that has a degree of similarity with the first representation layer vector that meets the specified requirements, wherein the first word vector matrix is stored in association with the information of the customer
  • the information of the customer includes at least the contact information of the customer;
  • the above product information is an introduction to the product, including text information such as the product name, use, and efficacy.
  • a method for vectorizing product information includes inputting each word in the product information into a preset first corpus dictionary to obtain a corresponding vector, and then combining the vectors of each word to form a first corresponding product information.
  • a product vector matrix is a dictionary of word vectors and words produced by a company that produces or sells the above-mentioned products, and the words and word vectors contained therein are common words in the field in which the above-mentioned products are located.
  • the process of making the first corpus dictionary includes: inputting text to be formed into a word vector into a DSSM (Deep Structured Semantic Models) model, calculating a word vector of the text through the DSSM model, and then writing the text and its corresponding The word vector is put into the first corpus dictionary.
  • the process of generating the first corpus dictionary includes: capturing related information of all products on a company website to form product information corresponding to each product; The words are extracted, and only one of each repeated word is retained; the remaining words are input into the DSSM model for calculation, and a word vector matrix of each word is obtained to obtain a first corpus dictionary.
  • the above-mentioned reverse recommendation model obtained based on the LSTM model training refers to that the reverse recommendation model is obtained by training the LSTM model.
  • the specific training process is as follows: 1 historical customer retrieval information and products purchased by the customer The information is vectorized through the above-mentioned first corpus dictionary and a corresponding set of word vector matrices is obtained; 2 the set of word vector matrices is divided into a training set and a test set; 3 the sample data of the test set is input into the LSTM model for Training, and then verify the trained LSTM model with the test set, if the verification is passed, the above reverse recommendation model is obtained.
  • the above training process belongs to supervised training, that is, historical customer retrieval information and product information purchased by customers are information with associated marks, and a backward recommendation model is trained.
  • the reverse recommendation model will output a first representation layer vector corresponding to the customer search information.
  • the above customer search information refers to the search information entered by the customer, which includes keywords extracted by the customer according to his or her desire to purchase. For example, if the customer wants to know about child insurance, then the keywords he enters generally include children, accidents, medical treatment, etc. After extracting these keywords, the word vector of each word is searched in the first corpus dictionary, and then the word vectors of each word are arranged in the order of the keywords to form the above-mentioned first word vector matrix.
  • step S3 it is to calculate the similarity between the first representation layer vector and each first word vector matrix in the customer database, and then the first word whose similarity with the first representation layer vector reaches the specified requirements Vector matrix extraction process.
  • the above specified requirement refers to a first word vector matrix whose similarity with the first representation layer vector reaches a specified threshold, which may be one or multiple.
  • the above-mentioned specified threshold will be appropriately set to facilitate finding a large number of customer groups. If only one with the highest similarity is found, it obviously does not meet the original intention of finding customers to recommend products.
  • the above contact methods generally include mobile phone numbers, email addresses, and other contact methods that can receive messages sent by others without adding friends.
  • a mobile phone number can be used to directly send text messages to customers, and the same is true for email addresses.
  • the customer information is associated with the first word vector matrix, so that the customer can be found backward through the first word vector matrix, and the first word vector matrix and the customer information are in a one-to-one mapping relationship.
  • the above similarity calculation can use Eucledian distance (Eucledian Distance, Manhattan Distance, Minkowski distance, or cosine similarity are calculated by an algorithm, so I won't go into details here.
  • the extracted first word vector matrix is a word vector matrix with a high degree of similarity to the first representation layer vector output by the reverse recommendation model, so its corresponding customer should have a relatively high purchase Desire and corresponding purchasing ability, so it is more efficient to send new insurance products to the corresponding customers and recommend resale.
  • step S4 of extracting the information of the customer corresponding to the first word vector matrix and recommending the product information to the customer according to the extracted contact method of the customer the method includes:
  • the first count ratio is used above the second count, and if the ratio is greater than a preset threshold, the reverse recommendation model is disabled.
  • an electronic archive is performed.
  • the customer's purchase situation can be automatically obtained from the electronic archive to know whether the customer has purchased the same product as the recommendation. That is, although a product is recommended to a customer for purchase, the customer does not necessarily purchase a product based on the recommendation. Record each customer's purchases and count them. When buying a different product from the recommendation, count it once. When buying the same product as the recommendation, count it once to get the first and second counts. Then calculate the ratio of the first count to the second count. If the ratio is large, the recommendation effect is poor. For example, the ratio is greater than 1, that is, the number of times of uneasy recommendation purchase is greater than the number of purchases according to recommendation.
  • the above recommendation model does not Ideal, so stop using it.
  • the third count purchased by all customers within a specified period of time may also be recorded, and then the first count or the second count may be compared with the third count, and based on the result of the ratio, it may be determined whether the inversion is required to be disabled.
  • the above specified time length refers to a longer period such as a quarter or a month, in order to obtain more data samples, and the obtained ratio is more available.
  • the step of extracting the information of the customer corresponding to the first word vector matrix and recommending the product information to the customer according to the extracted contact method of the customer includes:
  • the above sales data generally includes the quantity of the product sold; the above-mentioned customer distribution refers to the age distribution and regional distribution of the customers who purchased the product, that is, the customer distribution can be used to know that the product is purchased
  • the customer ’s situation for example, is that the product is a child's life insurance. Among the customers it purchases, there are more female customers aged 20-30, while there are fewer elderly customers aged 50-70 and fewer male customers. This can help customers understand What kind of people are buying this product, so that customers can choose whether to buy based on their own. Another example is that the product is a personal accident insurance. The customers who buy it are often in the winter in the north. It is related to the cold and easy to slip in the north. Getting the geographical distribution and time distribution is also helpful for new customers to choose whether to buy. Wait.
  • the introduction information, sales data and customer distribution of the above insurance products are presented to new customers, which facilitates new customers to make objective judgments and reduces the workload of product sales staff.
  • the above-mentioned step S41 of recommending the product information, sales data, and customer distribution to the customer according to the extracted contact information of the customer includes:
  • S412. Encapsulate the visual drawing and the product information into a document to form the recommendation information to recommend to the customer according to the extracted contact method of the customer.
  • the sales data and customer distribution are made into a visual drawing, which is more convenient for recommended customers to view.
  • the geographical distribution in the customer distribution is reflected on a map, so that customers can know at a glance Geographical differences, etc.
  • the sales data can be reflected in the form of a curve, such as the sales trend curve from the beginning of the product's sales to the present, or the monthly sales volume, which greatly improves the speed at which customers can view recommended information.
  • step S4 of extracting the information of the customer corresponding to the first word vector matrix and recommending the product information to the customer according to the extracted contact method of the customer the method includes:
  • the second word vector matrix is input to a recommendation model trained based on the LSTM model to learn, and a second representation layer vector corresponding to the second word vector matrix is output;
  • the new search information refers to the search information entered by the new customer, which includes keywords extracted by the new customer based on his own purchase desires. For example, if the customer wants to know about child insurance, the keywords entered by him Generally includes children, accidents, medical, etc. After extracting these keywords, the word vector of each word is found in the first corpus dictionary, and then the word vector of each word is arranged in the order of the keywords to form the above-mentioned First word vector matrix.
  • the keywords can be extracted by the sales staff based on the customer's oral content, and then manually entered into the computer equipment; or the customer's requirements are formed into text information and input into the keyword extraction together. Keywords are extracted from the model; it is also possible to use speech recognition technology to convert the customer's voice into text information, and then enter the text information into the keyword extraction model to extract keywords.
  • the recommendation model obtained based on the LSTM model training refers to that the recommendation model is obtained by training the LSTM model.
  • the specific training process is: 1 forming a plurality of historical customer retrieval information through the first corpus dictionary to form a plurality of The word vector matrix and the information of the product purchased by the customer are formed into a corresponding plurality of word vector matrices through a first corpus dictionary to form a set of word vector matrices.
  • the word vector matrix corresponding to the retrieval information of each customer is related to the customer.
  • the word vector matrix corresponding to the product information of the purchased product is related to each other; 2
  • the set of word vector matrices is divided into a training set and a test set; 3
  • the sample data of the test set is input to the LSTM model for training, and then the test set is trained on the training set.
  • the LSTM model is verified. If the verification is passed, the above recommended model is obtained.
  • the above training process belongs to supervised training, that is, historical customer retrieval information and customer purchased product information are information with associated tags.
  • the trained recommendation model is output when the second word vector matrix of new customer retrieval information is input.
  • step S9 because the second representation layer vector output in the above recommendation model is calculated, it is necessary to calculate the similarity between each representation layer vector in the product representation layer vector matrix of the trained product and find The third representation layer vector with the highest similarity can be calculated using the Eucledian distance. Distance, Manhattan Distance, Minkowski distance, or cosine similarity.
  • the specific similarity calculation formula is as follows:
  • yQ is the representation layer vector of the retrieved information
  • YD is the representation layer vector of the product information
  • TD / TQ is the length of the corresponding sentence
  • R is the similarity.
  • the product corresponding to the third representation layer vector is the most suitable product for new customers to purchase.
  • the product corresponding to the third representation layer vector is output as a product recommended to a new customer.
  • several third representation layer vectors with higher similarity to the second representation layer vector may also be selected, and the products corresponding to each third representation layer vector may be output in the order of their similarity. , Give suggestions for different purchase intensity.
  • the second representation layer vector output by the above recommendation model is closest to a third representation layer vector in the preset product representation layer vector matrix, and the third representation
  • the products corresponding to the layer vector are children's accident insurance, and the children's medical insurance corresponding to the third surface layer vector, which has a slightly lower similarity to the second representation layer vector corresponding to the above "children, insurance, accident", will also form a recommended insurance product.
  • the input search information is "Child, Insurance, Accident”
  • the method includes:
  • the second word vector matrix corresponding to the new search information of the new customer and the third product vector matrix are associatedly stored in a designated database;
  • the third model vector matrix and the second word vector matrix in the database are used to continue training the recommendation model to obtain a new recommendation model.
  • the new customer does not necessarily buy the recommended product, so the insurance product purchased by the new customer is recorded, and the product information of the product purchased by the new customer is vectorized
  • the third product vector matrix is obtained, and then the third product vector matrix and the second word vector matrix (the vector matrix corresponding to the customer's search information) are associated and stored in a designated database for later use as a training sample.
  • the above-mentioned designated data amount refers to a total data amount in which the third product vector matrix and the second word vector matrix are stored. When the amount of data reaches the specified threshold, it means that the number of samples meets the requirements, and then the above recommended model is trained.
  • the training method is the same as the method for training the recommendation model described above, and will not be repeated here.
  • the above method of vectorizing the information of the product purchased by the new customer to obtain a third product vector matrix is the same as the above-mentioned process of vectorizing the product information, and it will not be repeated here.
  • the intelligent recommendation method for product sales in the embodiment of the present application calculates a representation layer vector corresponding to the representation layer vector according to a vectorized matrix of product product information through a reverse recommendation model, and then searches for a similarity between the representation layer vector and a specified requirement.
  • the first word vector matrix, and the first word vector matrix is a vector matrix obtained by vectorizing the search information of the customer, so the contact information of the corresponding customer can be found according to the first word vector matrix, because the search information is actively input by the customer Information, so the customer corresponding to the found first word vector matrix should have a relatively high purchase desire and corresponding purchasing ability, so it is more efficient to send new products to the corresponding customers and recommend resale.
  • the present application further provides a smart recommendation device for product sales, including:
  • a first vectorization unit 10 configured to vectorize product information to obtain a first product vector matrix
  • An inverse learning unit 20 is configured to input the first product vector matrix into a backward recommendation model obtained based on the LSTM model training to output a first representation layer vector corresponding to the first product vector matrix;
  • the first A representation layer vector is a representation layer vector of a first word vector matrix corresponding to the retrieval information of the client;
  • the searching unit 30 is configured to search a preset customer database for the first word vector matrix that has a similarity with the first representation layer vector that meets a specified requirement, wherein the first word vector matrix and the client's Information is stored in the customer database in association, and the customer's information includes at least the customer's contact information;
  • the extraction recommendation unit 40 is configured to extract information of a customer corresponding to the first word vector matrix, and recommend the product information to the customer according to the extracted contact method of the customer.
  • the product information is an introduction to the product, including text information such as a product name, a use, and an effect.
  • a method for vectorizing product information includes inputting each word in the product information into a preset first corpus dictionary to obtain a corresponding vector, and then combining the vectors of each word to form a first corresponding product information.
  • a product vector matrix is a dictionary of word vectors and words produced by a company that produces or sells the above-mentioned products, and the words and word vectors contained therein are common words in the field in which the above-mentioned products are located.
  • the process of making the first corpus dictionary includes: inputting text to be formed into a word vector into a DSSM (Deep Structured Semantic Models) model, calculating a word vector of the text through the DSSM model, and then writing the text and its corresponding The word vector is put into the first corpus dictionary.
  • the process of generating the first corpus dictionary includes: capturing related information of all products on a company website to form product information corresponding to each product; The words are extracted, and only one of each repeated word is retained; the remaining words are input into the DSSM model for calculation, and a word vector matrix of each word is obtained to obtain a first corpus dictionary.
  • the inverse recommendation model obtained based on the LSTM model training refers to that the inverse recommendation model is obtained by training the LSTM model.
  • the specific training process is as follows: 1 retrieve historical customers and customers The purchased product information is vectorized through the above-mentioned first corpus dictionary, and a corresponding set of word vector matrices is obtained; 2 the set of word vector matrices is divided into a training set and a test set; 3 the sample data of the test set is input into the LSTM The model is trained, and then the test set is used to verify the trained LSTM model. If the verification is passed, the above reverse recommendation model is obtained.
  • the above training process belongs to supervised training, that is, historical customer retrieval information and product information purchased by customers are information with associated marks, and a backward recommendation model is trained.
  • the reverse recommendation model will output a first representation layer vector corresponding to the customer search information.
  • the above customer search information refers to the search information entered by the customer, which includes keywords extracted by the customer according to his or her desire to purchase. For example, if the customer wants to know about child insurance, then the keywords he enters generally include children, accidents, medical treatment, etc. After extracting these keywords, the word vector of each word is searched in the first corpus dictionary, and then the word vectors of each word are arranged in the order of the keywords to form the above-mentioned first word vector matrix.
  • the search unit 30 it is to calculate the similarity between the first representation layer vector and each first word vector matrix in the customer database, and then the first similarity with the first representation layer vector reaches the specified requirements.
  • the process of extracting the word vector matrix refers to a first word vector matrix whose similarity with the first representation layer vector reaches a specified threshold, which may be one or multiple.
  • the above-mentioned specified threshold will be appropriately set to facilitate finding a large number of customer groups. If only one with the highest similarity is found, it obviously does not meet the original intention of finding customers to recommend products.
  • the above contact methods generally include mobile phone numbers, email addresses, and other contact methods that can receive messages sent by others without adding friends.
  • a mobile phone number can be used to directly send text messages to customers, and the same is true for email addresses.
  • the customer information is associated with the first word vector matrix, so that the customer can be found backward through the first word vector matrix, and the first word vector matrix and the customer information are in a one-to-one mapping relationship.
  • the above similarity calculation can use Eucledian distance (Eucledian Distance, Manhattan Distance, Minkowski distance, or cosine similarity are calculated by an algorithm, so I won't go into details here.
  • the extracted first word vector matrix is a word vector matrix with a high degree of similarity to the first representation layer vector output by the reverse recommendation model, so its corresponding customer should have a relatively high The desire to purchase and the corresponding purchasing ability, so it is more efficient to send new insurance products to the corresponding customers.
  • the intelligent recommendation device for product sales described above further includes:
  • An acquisition judgment unit configured to acquire a purchased product purchased by the customer, and determine whether the purchased product is the same as the product
  • a counting unit configured to add one to the count of different products purchased if the purchased product is not the same as the product, and add one to the count of the same product if they are the same; To get the second count
  • the comparison deactivation unit is configured to use a first count ratio over a second count at a time node of a specified time length, and if the ratio is greater than a preset threshold value, disable the reverse recommendation model.
  • an electronic archive is performed, and the purchase situation of the customer can be automatically obtained from the electronic archive to know whether the customer has purchased the same product as the recommendation. That is, although a product is recommended to a customer for purchase, the customer does not necessarily purchase a product based on the recommendation. Record each customer's purchases and count them. When buying a different product from the recommendation, count it once. When buying the same product as the recommendation, count it once to get the first and second counts. Then calculate the ratio of the first count to the second count. If the ratio is large, the recommendation effect is poor. For example, the ratio is greater than 1, that is, the number of times of uneasy recommendation purchase is greater than the number of purchases according to recommendation.
  • the above recommendation model does not Ideal, so stop using it.
  • the third count purchased by all customers within a specified period of time may also be recorded, and then the first count or the second count may be compared with the third count, and based on the result of the ratio, it may be determined whether the inversion is required to be disabled Recommended model.
  • the above specified time length refers to a longer period such as a quarter or a month, in order to obtain more data samples, and the obtained ratio is more available.
  • the above-mentioned extraction recommendation unit 40 includes:
  • An acquisition module for acquiring sales data of the product and a distribution of customers who purchase the product
  • a recommendation module is configured to recommend information of the product information, sales data, and customer distribution to recommend the customer according to the extracted contact information of the customer.
  • the above sales data generally includes the quantity of the product sold; the above-mentioned customer distribution refers to the age distribution, regional distribution, etc. of the customers who purchased the product, that is, through the customer distribution, it is possible to know the customers who purchased the product.
  • the product is children's life insurance.
  • the customers it purchases there are more female customers aged 20-30, and slightly fewer elderly customers aged 50-70. Fewer male customers, etc., can make customers understand the purchase of the product. What kind of crowd is the crowd, so that customers can choose whether to buy based on their own.
  • the product is a personal accident insurance.
  • the customers who buy it are often in the winter in the north. It is related to the cold and easy to slip in the north. Getting the geographical distribution and time distribution is also helpful for new customers to choose whether to buy. Wait.
  • the introduction information, sales data and customer distribution of the above insurance products are presented to new customers, which facilitates new customers to make objective judgments and reduces the workload of product sales staff.
  • the above recommendation module includes:
  • a visualization subunit configured to make the sales data and customer distribution into a visual drawing
  • a recommendation sub-module is configured to encapsulate the visualization drawing and the information of the product into a document to form the recommendation information to recommend to the customer according to the extracted contact information of the customer.
  • the sales data and customer distribution are made into a visual drawing, which is more convenient for recommended customers to view.
  • the geographical distribution in the customer distribution is reflected on a map, so that customers can know the regional differences at a glance.
  • the sales data can be reflected in the form of a curve, such as the sales trend curve from the beginning of the product's sales to the present, or the monthly sales volume, which greatly improves the speed at which customers can view recommended information.
  • the smart recommendation device for product sales includes:
  • An acquisition vector unit configured to acquire new search information of a new customer, and vectorize the new search information to obtain a corresponding second word vector matrix
  • a recommendation learning unit configured to input a second word vector matrix into a recommendation model trained based on the LSTM model to learn, and output a second representation layer vector corresponding to the second word vector matrix;
  • the recommendation unit is configured to output a product corresponding to the third representation layer vector as a product recommended to a new customer.
  • the above-mentioned new search information refers to search information input by new customers, which includes keywords extracted by new customers based on their own purchasing desires.
  • the words generally include children, accidents, medical, etc.
  • the keywords can be extracted by the sales staff based on the customer's oral content, and then manually entered into the computer equipment; or the customer's requirements are formed into text information and input into the keyword extraction together. Keywords are extracted from the model; it is also possible to use speech recognition technology to convert the customer's voice into text information, and then enter the text information into the keyword extraction model to extract keywords.
  • the above recommendation model obtained based on the LSTM model training refers to the recommendation model obtained through the training of the LSTM model.
  • the specific training process is as follows: 1
  • the historical customer retrieval information is formed through the above-mentioned first corpus dictionary.
  • the above training process belongs to supervised training, that is, historical customer retrieval information and customer purchased product information are information with associated tags.
  • the trained recommendation model is output when the second word vector matrix of new customer retrieval information is input.
  • yQ is the representation layer vector of the retrieved information
  • YD is the representation layer vector of the product information
  • TD / TQ is the length of the corresponding sentence
  • R is the similarity.
  • the product corresponding to the third representation layer vector is the most suitable product for new customers to purchase.
  • the product corresponding to the third representation layer vector is output as a product recommended to a new customer.
  • several third representation layer vectors with higher similarity to the second representation layer vector may also be selected, and the products corresponding to each third representation layer vector may be output in the order of their similarity. , Give suggestions for different purchase intensity.
  • the second representation layer vector output by the above recommendation model is closest to a third representation layer vector in the preset product representation layer vector matrix, and the third representation
  • the products corresponding to the layer vector are children's accident insurance, and the children's medical insurance corresponding to the third surface layer vector, which has a slightly lower similarity to the second representation layer vector corresponding to the above "children, insurance, accident", will also form a recommended insurance product.
  • the input search information is "Child, Insurance, Accident”
  • the intelligent recommendation device for the above product sales further includes:
  • a recording unit configured to record products purchased by the new customer, and search information of the new customer
  • a second vectorization unit configured to vectorize information about products purchased by the new customer to obtain a third product vector matrix
  • An association unit configured to associate the second word vector matrix corresponding to the new search information of the new customer and the third product vector matrix to a specified database
  • An updating unit configured to continue training the recommendation model by using all the third product vector matrices and the second word vector matrix in the database when the amount of data in the database reaches a preset threshold to obtain a new Recommended model.
  • a product is recommended to a new customer
  • the new customer does not necessarily purchase the recommended product, so the insurance product purchased by the new customer is recorded, and the product information of the product purchased by the new customer is vectorized to a third.
  • Product vector matrix and then store the third product vector matrix and the second word vector matrix (the vector matrix corresponding to the customer's retrieval information) in the specified database for later use as training samples.
  • the above-mentioned designated data amount refers to a total data amount in which the third product vector matrix and the second word vector matrix are stored.
  • the amount of data reaches the specified threshold, it means that the number of samples meets the requirements, and then the above recommended model is trained. Using the sample data in the database to continue training the recommended model can improve the accuracy of the above recommended model.
  • the training method is the same as the method for training the recommendation model described above, and will not be repeated here.
  • the above method of vectorizing the information of the product purchased by the new customer to obtain a third product vector matrix is the same as the above-mentioned process of vectorizing the product information, and it will not be repeated here.
  • the intelligent recommendation device for product sales in the embodiment of the present application calculates a representation layer vector corresponding to the representation layer vector according to a vectorized matrix of product product information through a reverse recommendation model, and then searches for a similarity between the representation layer vector and a specified requirement.
  • the first word vector matrix, and the first word vector matrix is a vector matrix obtained by vectorizing the search information of the customer, so the contact information of the corresponding customer can be found according to the first word vector matrix, because the search information is actively input by the customer Information, so the customer corresponding to the found first word vector matrix should have a relatively high purchase desire and corresponding purchasing ability, so it is more efficient to send new products to the corresponding customers and recommend resale.
  • an embodiment of the present invention further provides a computer device.
  • the computer device may be a server, and its internal structure may be as shown in FIG. 3.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the computer design processor is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the memory provides an environment for operating systems and computer-readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used to store reverse recommendation models and recommendation models.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement the processes of the embodiments of the methods described above.
  • An embodiment of the present invention also provides a computer non-volatile readable storage medium, which stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the processes of the embodiments of the foregoing methods are implemented.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Human Resources & Organizations (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请揭示了一种产品销售的智能推荐方法、装置、计算机设备和存储介质，根据产品的产品信息的向量化矩阵，通过反向推荐模型计算出与其相对应的表示层向量，然后查找与该表示层向量相似度达到指定要求的第一词向量矩阵，根据第一词向量矩阵查找到对应客户的联系方式，将新的产品发送给对应的客户，推荐转销售的效率更高。

Description

产品销售的智能推荐方法、装置、计算机设备和存储介质
本申请要求于2018年9月5日提交中国专利局、申请号为2018110329333，申请名称为“产品销售的智能推荐方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及到计算机领域，特别是涉及到一种产品销售的智能推荐方法、装置、计算机设备和存储介质。
 背景技术
保险、投资理财的时候，会有相关的系统进行统计与计算，以生成推荐信息给客户买哪些保险或理财产品等。现有的推荐系统或是基于内容推荐、或是基于用户推荐，但是当新的保险产品推出时，不能有效进行推荐，而有些热门保险产品会被过度推荐。
 技术问题
本申请的主要目的为提供一种可以有效地智能推荐销售产品的产品销售的智能推荐方法、装置、计算机设备和存储介质。
 技术解决方案
为了实现上述发明目的，本申请提出一种产品销售的智能推荐方法，包括：
将产品信息向量化，得到第一产品向量矩阵；
将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
本申请还提供一种产品销售的智能推荐装置，包括：
第一向量化单元，用于将产品信息向量化，得到第一产品向量矩阵；
反向学习单元，用于将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
查找单元，用于到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
提取推荐单元，用于提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
本申请还提供一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现上述任一项所述方法的步骤。
本申请还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现上述任一项所述的方法的步骤。
 有益效果
本申请的产品销售的智能推荐方法、装置、计算机设备和存储介质，根据产品的产品信息的向量化矩阵，通过反向推荐模型计算出与其相对应的表示层向量，然后查找与该表示层向量相似度达到指定要求的第一词向量矩阵，而该第一词向量矩阵是利用客户的检索信息向量化得到的向量矩阵，所以可以根据第一词向量矩阵查找到对应客户的联系方式，因为检索信息是客户主动输入的信息，所以查找到的第一词向量矩阵对应的客户应该具有相对较高的购买欲望，以及相应的购买能力，所以将新的产品发送给对应的客户，推荐转销售的效率。更高。
 附图说明
图1为本申请一实施例的产品销售的智能推荐方法的流程示意图；
图2 为本申请一实施例的产品销售的智能推荐装置的结构示意框图；
图3 为本申请一实施例的计算机设备的结构示意框图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
 本发明的最佳实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
参照图1，本申请实施例首先提供一种产品销售的智能推荐方法，包括步骤：
S1、将产品信息向量化，得到第一产品向量矩阵；
S2、将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；其中，所述第一表示层向量是对应所述客户的检索信息的第一词向量矩阵的表示层向量；
S3、到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
S4、提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
如上述步骤S1所述，上述产品信息即为产品的介绍，包括产品名称、用途、功效等文字信息。将产品信息向量化的方法包括，将产品信息中的每一个字输入到预设的第一语料字典中，以得到对应的向量，然后每一个字的向量组合在一起，形成对应产品信息的第一产品向量矩阵。上述的第一语料字典是生产或销售上述产品的公司制作的词向量与文字一一映射的字典，其中包含的文字和词向量是其上述产品所处领域中常见的字。第一语料词典的制作过程包括：将待形成词向量的文字输入到DSSM（Deep Structured Semantic Models，深层结构化语义模型）模型中，通过DSSM模型计算出文字的词向量，然后将文字及其对应的词向量放入到第一语料字典中。本申请实施例中，在一个实施例中，上述第一语料字典的生成过程包括：抓取公司网站上的全部产品的相关信息，形成各产品对应的产品信息；将各产品信息中重复出现的字提取出来，每一个重复的字只保留一个；将剩余的字输入到DSSM模型中计算，得出每一个字的词向量矩阵，以得到第一语料字典。
如上述步骤S2所述，上述基于LSTM模型训练得到的反向推荐模型是指，反向推荐模型通过LSTM模型训练而得，具体的训练过程为：①将历史的客户检索信息和客户购买的产品信息通过上述的第一语料字典进行向量化，并得到对应的词向量矩阵的集合；②将该词向量矩阵的集合分成训练集和测试集；③将测试集的样本数据输入到LSTM模型中进行训练，然后将测试集对训练后的LSTM模型进行验证，如果验证通过，则得到上述的反向推荐模型。上述训练的过程属于监督训练，即历史的客户检索信息和客户购买的产品信息是带有关联标记的信息，训练出的反向推荐模型。当输入产品信息的第一产品向量矩阵到反向推荐模型后，反向推荐模型会输出一个与客户检索信息对应的第一表示层向量。上述客户检索信息是指客户输入的检索信息，其包括客户根据自身购买欲望而提炼出的关键词，如，客户想要了解儿童保险，那么其输入的关键词一般包括儿童、意外、医疗等，将这些关键词提取出来之后，在上述第一语料字典中查找每一个字的词向量，然后将每一个字的词向量按照关键词的顺序排列，形成上述的第一词向量矩阵。
如上述步骤S3所述，即为将第一表示层向量分别与客户数据库中的各第一词向量矩阵进行像似度计算，然后将与第一表示层向量相似度达到指定要求的第一词向量矩阵提取出来的过程。上述指定要求是指与第一表示层向量相似度达到指定阈值的第一词向量矩阵，其可能是一个，也可能是多个。本实施例中，上述指定阈值会适当设定，以便于查找到大量的客户群体，如只查找一个相似度最高的，显然不符合查找客户以推荐产品的初衷。上述联系方式一般包括手机号码、邮箱等无需添加好友，既可以接收他人发送的消息的联系方式，比如，通过手机号码可以直接发送短信给客户，邮箱同样如此。将客户的信息与第一词向量矩阵关联，是为了可以通过第一词向量矩阵反向查找到客户，第一词向量矩阵与客户的信息是一对一的映射关系。上述相似度的计算可以使用欧几里得距离（Eucledian Distance）、曼哈顿距离（Manhattan Distance）、明可夫斯基距离（Minkowski distance）或者余弦相似度中一种算法进行计算，再此不在赘述。
如上述步骤S4所述，上述提取出的第一词向量矩阵是与反向推荐模型输出的第一表示层向量相似度较高的词向量矩阵，所以其对应的客户应该具有相对较高的购买欲望，以及相应的购买能力，所以将新的保险产品发送给对应的客户，推荐转销售的效率更高。
在一个实施例中，上述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤S4之后，包括：
S5、获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
S6、若不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数
S7、在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
如上述步骤S5至S7所述，客户购买产品后会进行电子存档，可以从电子存档中自动获取客户的购买情况，以了解客户是否购买了与推荐相同的产品。即，虽然推荐给客户购买产品，但是客户并不一定会根据推荐而购买产品。记录每一次客户的购买情况，并进行计数，当购买与推荐不同的产品时，计数一次，购买与推荐相同的产品时计数一次，即得到上述的第一计数和第二计数。然后计算第一计数比上第二计数的比值，如果比值较大，说明推荐效果较差，比如，比值大于1，即不安推荐购买的次数大于按照推荐购买的次数，显然，上述推荐模型并不理想，所以停止使用。在其它实施例中，还可以记录在指定时间长度内的全部客户购买的第三计数，然后将第一计数或第二计数比上第三计数，根据比值结果，判断是否需要停用上述反向推荐模型。上述的指定时间长度是指一个季度或者一个月等较长时间，以便于得到较多的数据样本，得出的比值的可用性更高。
在一个实施例中，上述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤S4，包括：
S41、获取所述产品的售卖数据，以及购买所述产品的客户分布；
S42、将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户。
如上述步骤S41和S42所述，上述的售卖数据一般包括该产品售卖的数量等；上述客户分布是指购买该产品的客户的年龄分布、区域分布等，即通过客户分布，可以知道购买该产品的客户的情况，如，产品为儿童人身保险，其购买的客户中，20-30岁的女性客户较多，而50-70岁的老人客户略少，男性客户更少等，可以使客户了解购买该产品的人群是什么样的人群，以便于客户结合自身选择是否购买。又比如，产品为人身意外险，其购买的客户在北方冬天的时候较多，其与北方天寒地冻容易滑倒相关，得到此地域分布和时间分布，也有利于帮助新客户选择是否购买等。将上述的保险产品的介绍信息、售卖数据和客户分布一起呈现给新客户，方便新客户进行客观的判断，也减少产品的售卖人员的工作量。
在一个实施例中，上述将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户的步骤S41包括：
S411、将所述售卖数据和客户分布制作成可视化附图；
S412、将所述可视化附图以及所述产品的信息封装到一篇文档中形成所述推荐信息按照提取出的所述客户的联系方式推荐给客户。
如上述步骤S411和S412所述，将售卖数据和客户分布制成可视化附图，更方便被推荐的客户查看，比如客户分布中的地域分布是体现在一张地图上，使客户可以一目了然地知道地域差异等。而售卖数据可以以曲线的形式体现，如从产品开始售卖开始至今的售卖走势曲线，或者每个月的售卖量等，大大地提高客户查看推荐信息的速度。
在一个实施例中，上述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤S4之后，包括：
S7、获取新客户的新检索信息，并将所述新检索信息向量化得到对应的所述第二词向量矩阵；
S8、将第二词向量矩阵输入到基于LSTM模型训练得到的推荐模型中学习，输出对应所述第二词向量矩阵的第二表示层向量；
S9、到训练好的对应产品的产品表示层向量矩阵中查找与所述第二表示层向量相似度最高的第三表示层向量；
S10、将第三表示层向量对应的产品输出，作为推荐给新客户的产品。
如上述步骤S7所述，上述新检索信息是指新客户输入的检索信息，其包括新客户根据自身购买欲望而提炼出的关键词，如，客户想要了解儿童保险，那么其输入的关键词一般包括儿童、意外、医疗等，将这些关键词提取出来之后，在上述第一语料字典中查找每一个字的词向量，然后将每一个字的词向量按照关键词的顺序排列，形成上述的第一词向量矩阵。本步骤中，关键词的提取可以是售卖产品的工作人员根据客户的口述内容，人为提取出来的，然后手动输入计算机设备中；也可以是将客户的要求形成文本信息，一起输入到关键词提取模型中提取关键词；还可以是利用语音识别技术将客户的语音转化成文本信息，然后将文本信息输入到关键词提取模型中提取关键词等。
如上述步骤S8所述，上述基于LSTM模型训练得到的推荐模型，是指推荐模型通过LSTM模型训练而得，具体的训练过程为：①将历史的客户检索信息通过上述第一语料字典形成多个词向量矩阵，以及将客户购买的产品的信息通过第一语料字典形成对应的多个词向量矩阵，形成词向量矩阵的集合，其中，每一位客户的检索信息对应的词向量矩阵与该客户购买的产品的产品信息对应的词向量矩阵相互关联；②将词向量矩阵的集合分成训练集和测试集；③将测试集的样本数据输入到LSTM模型中进行训练，然后将测试集对训练后的LSTM模型进行验证，如果验证通过，则得到上述的推荐模型。上述训练的过程属于监督训练，即历史的客户检索信息和客户购买的产品信息是带有关联标记的信息，训练出的推荐模型，当输入新客户检索信息的第二词向量矩阵后，会输出一个关于产品信息对应的第二表示层向量。
如上述步骤S9所述，因为上述推荐模型中输出的第二表示层向量是推算出来的，所以需要计算与训练好的产品的产品表示层向量矩阵中的每一个表示层向量的相似度，寻找相似度最高的第三表示层向量，相似度的计算可以使用欧几里得距离（Eucledian Distance）、曼哈顿距离（Manhattan Distance）、明可夫斯基距离（Minkowski distance）或者余弦相似度中一种算法进行计算。本实施例中，具体的相似度计算公式如下：
Figure dest_path_image002
其中， yQ 为检索信息的表示层向量,YD为产品信息的表示层向量,TD/TQ为对应句子的长度，R 为相似度。在其它的实施例中，无论有没有与第二表示层向量相同的第三表示层向量，都会进行上述的相似度计算，并将计算结果进行由高到低的排名。
如上述步骤S10所述，因为上述第三表示层向量与推荐模型输出的第二表示层向量的相似度最高，所以，第三表示层向量对应的产品是最适合新客户购买的产品，因此将该第三表示层向量对应的产品输出，作为推荐给新客户的产品。在其它实施例中，还可以将与第二表示层向量相似度较高的几个第三表示层向量选取出来，并将各第三表示层向量对应的产品输出，按照其相似度高低的顺序，给出不同购买强度的建议。比如，输入的检索信息为“儿童、保险、意外”，那么上述推荐模型输出的第二表示层向量与预设的产品表示层向量矩阵中的一个第三表示层向量最接近，该第三表示层向量对应的产品为儿童意外保险，而与上述“儿童、保险、意外”对应的第二表示层向量相似度略低一点的第三表层向量对应的儿童医疗保险同样会形成推荐的保险产品推出，以提高客户的选择性。
在一个实施例中，上述将第三表示层向量对应的产品输出，作为推荐给所述新客户的产品的步骤S10之后，包括：
S11、记录所述新客户购买的产品，以及所述新客户的检索信息；
S12、将所述新客户购买的产品的信息进行向量化得第三产品向量矩阵；
S13、将上所述新客户的新检索信息对应的第二词向量矩阵和所述第三产品向量矩阵关联地保存到指定的数据库中；
S14、当所述数据库中的数据量达到预设的阈值后，利用数据库中的全部第三产品向量矩阵和第二词向量矩阵对所述推荐模型进行继续训练，得到新的所述推荐模型。
如上述步骤S11至S14所述，虽然给新客户推荐了产品，但是新客户并不一定会购买推荐的产品，所以记录该新客户购买的保险产品，将新客户购买的产品的产品信息向量化得到第三产品向量矩阵，然后将第三产品向量矩阵与第二词向量矩阵（客户的检索信息对应的向量矩阵）关联存储到指定的数据库中，以便后期作为训练样本使用。上述指定数据量是指存储上述第三产品向量矩阵与第二词向量矩阵的总的数据量。当数据量达到指定阈值，则说明样本数量达到要求，然后对上述的推荐模型进行训练，使用数据库中的样本数据继续对推荐模型进行训练，可以提高上述推荐模型的准确性。其训练方法与上述训练推荐模型的方法相同，在此不在赘述。上述对新客户购买的产品的信息进行向量化得第三产品向量矩阵的方法与上述将产品信息向量化的过程相同，在此同样不在赘述。
本申请实施例的产品销售的智能推荐方法，根据产品的产品信息的向量化矩阵，通过反向推荐模型计算出与其相对应的表示层向量，然后查找与该表示层向量相似度达到指定要求的第一词向量矩阵，而该第一词向量矩阵是利用客户的检索信息向量化得到的向量矩阵，所以可以根据第一词向量矩阵查找到对应客户的联系方式，因为检索信息是客户主动输入的信息，所以查找到的第一词向量矩阵对应的客户应该具有相对较高的购买欲望，以及相应的购买能力，所以将新的产品发送给对应的客户，推荐转销售的效率更高。
参照图2，本申请还提供一种产品销售的智能推荐装置，包括：
第一向量化单元10，用于将产品信息向量化，得到第一产品向量矩阵；
反向学习单元20，用于将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
查找单元30，用于到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
提取推荐单元40，用于提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
如上述第一向量化单元10所述，上述产品信息即为产品的介绍，包括产品名称、用途、功效等文字信息。将产品信息向量化的方法包括，将产品信息中的每一个字输入到预设的第一语料字典中，以得到对应的向量，然后每一个字的向量组合在一起，形成对应产品信息的第一产品向量矩阵。上述的第一语料字典是生产或销售上述产品的公司制作的词向量与文字一一映射的字典，其中包含的文字和词向量是其上述产品所处领域中常见的字。第一语料词典的制作过程包括：将待形成词向量的文字输入到DSSM（Deep Structured Semantic Models，深层结构化语义模型）模型中，通过DSSM模型计算出文字的词向量，然后将文字及其对应的词向量放入到第一语料字典中。本申请实施例中，在一个实施例中，上述第一语料字典的生成过程包括：抓取公司网站上的全部产品的相关信息，形成各产品对应的产品信息；将各产品信息中重复出现的字提取出来，每一个重复的字只保留一个；将剩余的字输入到DSSM模型中计算，得出每一个字的词向量矩阵，以得到第一语料字典。
如上述反向学习单元20所述，上述基于LSTM模型训练得到的反向推荐模型是指，反向推荐模型通过LSTM模型训练而得，具体的训练过程为：①将历史的客户检索信息和客户购买的产品信息通过上述的第一语料字典进行向量化，并得到对应的词向量矩阵的集合；②将该词向量矩阵的集合分成训练集和测试集；③将测试集的样本数据输入到LSTM模型中进行训练，然后将测试集对训练后的LSTM模型进行验证，如果验证通过，则得到上述的反向推荐模型。上述训练的过程属于监督训练，即历史的客户检索信息和客户购买的产品信息是带有关联标记的信息，训练出的反向推荐模型。当输入产品信息的第一产品向量矩阵到反向推荐模型后，反向推荐模型会输出一个与客户检索信息对应的第一表示层向量。上述客户检索信息是指客户输入的检索信息，其包括客户根据自身购买欲望而提炼出的关键词，如，客户想要了解儿童保险，那么其输入的关键词一般包括儿童、意外、医疗等，将这些关键词提取出来之后，在上述第一语料字典中查找每一个字的词向量，然后将每一个字的词向量按照关键词的顺序排列，形成上述的第一词向量矩阵。
如上述查找单元30所述，即为将第一表示层向量分别与客户数据库中的各第一词向量矩阵进行像似度计算，然后将与第一表示层向量相似度达到指定要求的第一词向量矩阵提取出来的过程。上述指定要求是指与第一表示层向量相似度达到指定阈值的第一词向量矩阵，其可能是一个，也可能是多个。本实施例中，上述指定阈值会适当设定，以便于查找到大量的客户群体，如只查找一个相似度最高的，显然不符合查找客户以推荐产品的初衷。上述联系方式一般包括手机号码、邮箱等无需添加好友，既可以接收他人发送的消息的联系方式，比如，通过手机号码可以直接发送短信给客户，邮箱同样如此。将客户的信息与第一词向量矩阵关联，是为了可以通过第一词向量矩阵反向查找到客户，第一词向量矩阵与客户的信息是一对一的映射关系。上述相似度的计算可以使用欧几里得距离（Eucledian Distance）、曼哈顿距离（Manhattan Distance）、明可夫斯基距离（Minkowski distance）或者余弦相似度中一种算法进行计算，再此不在赘述。
如上述提取推荐单元40所述，上述提取出的第一词向量矩阵是与反向推荐模型输出的第一表示层向量相似度较高的词向量矩阵，所以其对应的客户应该具有相对较高的购买欲望，以及相应的购买能力，所以将新的保险产品发送给对应的客户，推荐转销售的效率更高。
在一个实施例中，上述产品销售的智能推荐装置，还包括：
获取判断单元，用于获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
计数单元，用于若所述购买产品与所述产品不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数
比较停用单元，用于在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
在本实施例中，上述客户购买产品后会进行电子存档，可以从电子存档中自动获取客户的购买情况，以了解客户是否购买了与推荐相同的产品。即，虽然推荐给客户购买产品，但是客户并不一定会根据推荐而购买产品。记录每一次客户的购买情况，并进行计数，当购买与推荐不同的产品时，计数一次，购买与推荐相同的产品时计数一次，即得到上述的第一计数和第二计数。然后计算第一计数比上第二计数的比值，如果比值较大，说明推荐效果较差，比如，比值大于1，即不安推荐购买的次数大于按照推荐购买的次数，显然，上述推荐模型并不理想，所以停止使用。在其它实施例中，还可以记录在指定时间长度内的全部客户购买的第三计数，然后将第一计数或第二计数比上第三计数，根据比值结果，判断是否需要停用上述反向推荐模型。上述的指定时间长度是指一个季度或者一个月等较长时间，以便于得到较多的数据样本，得出的比值的可用性更高。
在一个实施例中，上述提取推荐单元40，包括：
获取模块，用于获取所述产品的售卖数据，以及购买所述产品的客户分布；
推荐模块，用于将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户。
在本实施例中，上述的售卖数据一般包括该产品售卖的数量等；上述客户分布是指购买该产品的客户的年龄分布、区域分布等，即通过客户分布，可以知道购买该产品的客户的情况，如，产品为儿童人身保险，其购买的客户中，20-30岁的女性客户较多，而50-70岁的老人客户略少，男性客户更少等，可以使客户了解购买该产品的人群是什么样的人群，以便于客户结合自身选择是否购买。又比如，产品为人身意外险，其购买的客户在北方冬天的时候较多，其与北方天寒地冻容易滑倒相关，得到此地域分布和时间分布，也有利于帮助新客户选择是否购买等。将上述的保险产品的介绍信息、售卖数据和客户分布一起呈现给新客户，方便新客户进行客观的判断，也减少产品的售卖人员的工作量。
在一个实施例中，上述推荐模块，包括：
可视化子单元，用于将所述售卖数据和客户分布制作成可视化附图；
推荐子模块，用于将所述可视化附图以及所述产品的信息封装到一篇文档中形成所述推荐信息按照提取出的所述客户的联系方式推荐给客户。
本实施例中，将售卖数据和客户分布制成可视化附图，更方便被推荐的客户查看，比如客户分布中的地域分布是体现在一张地图上，使客户可以一目了然地知道地域差异等。而售卖数据可以以曲线的形式体现，如从产品开始售卖开始至今的售卖走势曲线，或者每个月的售卖量等，大大地提高客户查看推荐信息的速度。
在一个实施例中，上述产品销售的智能推荐装置，包括：
获取向量单元，用于获取新客户的新检索信息，并将所述新检索信息向量化得到对应的所述第二词向量矩阵；
推荐学习单元，用于将第二词向量矩阵输入到基于LSTM模型训练得到的推荐模型中学习，输出对应所述第二词向量矩阵的第二表示层向量；
像似查找单元，用于到训练好的对应产品的产品表示层向量矩阵中查找与所述第二表示层向量相似度最高的第三表示层向量；
推荐单元，用于将第三表示层向量对应的产品输出，作为推荐给新客户的产品。
如上述获取向量单元所述，上述新检索信息是指新客户输入的检索信息，其包括新客户根据自身购买欲望而提炼出的关键词，如，客户想要了解儿童保险，那么其输入的关键词一般包括儿童、意外、医疗等，将这些关键词提取出来之后，在上述第一语料字典中查找每一个字的词向量，然后将每一个字的词向量按照关键词的顺序排列，形成上述的第一词向量矩阵。本步骤中，关键词的提取可以是售卖产品的工作人员根据客户的口述内容，人为提取出来的，然后手动输入计算机设备中；也可以是将客户的要求形成文本信息，一起输入到关键词提取模型中提取关键词；还可以是利用语音识别技术将客户的语音转化成文本信息，然后将文本信息输入到关键词提取模型中提取关键词等。
如上述推荐学习单元所述，上述基于LSTM模型训练得到的推荐模型，是指推荐模型通过LSTM模型训练而得，具体的训练过程为：①将历史的客户检索信息通过上述第一语料字典形成多个词向量矩阵，以及将客户购买的产品的信息通过第一语料字典形成对应的多个词向量矩阵，形成词向量矩阵的集合，其中，每一位客户的检索信息对应的词向量矩阵与该客户购买的产品的产品信息对应的词向量矩阵相互关联；②将词向量矩阵的集合分成训练集和测试集；③将测试集的样本数据输入到LSTM模型中进行训练，然后将测试集对训练后的LSTM模型进行验证，如果验证通过，则得到上述的推荐模型。上述训练的过程属于监督训练，即历史的客户检索信息和客户购买的产品信息是带有关联标记的信息，训练出的推荐模型，当输入新客户检索信息的第二词向量矩阵后，会输出一个关于产品信息对应的第二表示层向量。
如上述像似查找单元所述，因为上述推荐模型中输出的第二表示层向量是推算出来的，所以需要计算与训练好的产品的产品表示层向量矩阵中的每一个表示层向量的相似度，寻找相似度最高的第三表示层向量，相似度的计算可以使用欧几里得距离（Eucledian Distance）、曼哈顿距离（Manhattan Distance）、明可夫斯基距离（Minkowski distance）或者余弦相似度中一种算法进行计算。本实施例中，具体的相似度计算公式如下：
Figure WO-DOC-FIGURE-1
其中， yQ 为检索信息的表示层向量,YD为产品信息的表示层向量,TD/TQ为对应句子的长度，R 为相似度。在其它的实施例中，无论有没有与第二表示层向量相同的第三表示层向量，都会进行上述的相似度计算，并将计算结果进行由高到低的排名。
如上述推荐单元所述，因为上述第三表示层向量与推荐模型输出的第二表示层向量的相似度最高，所以，第三表示层向量对应的产品是最适合新客户购买的产品，因此将该第三表示层向量对应的产品输出，作为推荐给新客户的产品。在其它实施例中，还可以将与第二表示层向量相似度较高的几个第三表示层向量选取出来，并将各第三表示层向量对应的产品输出，按照其相似度高低的顺序，给出不同购买强度的建议。比如，输入的检索信息为“儿童、保险、意外”，那么上述推荐模型输出的第二表示层向量与预设的产品表示层向量矩阵中的一个第三表示层向量最接近，该第三表示层向量对应的产品为儿童意外保险，而与上述“儿童、保险、意外”对应的第二表示层向量相似度略低一点的第三表层向量对应的儿童医疗保险同样会形成推荐的保险产品推出，以提高客户的选择性。
进一步地，上述产品销售的智能推荐装置，还包括：
记录单元，用于记录所述新客户购买的产品，以及所述新客户的检索信息；
第二向量化单元，用于将所述新客户购买的产品的信息进行向量化得第三产品向量矩阵；
关联单元，用于将上所述新客户的新检索信息对应的第二词向量矩阵和所述第三产品向量矩阵关联地保存到指定的数据库中；
更新单元，用于当所述数据库中的数据量达到预设的阈值后，利用数据库中的全部第三产品向量矩阵和第二词向量矩阵对所述推荐模型进行继续训练，得到新的所述推荐模型。
在本实施例中，虽然给新客户推荐了产品，但是新客户并不一定会购买推荐的产品，所以记录该新客户购买的保险产品，将新客户购买的产品的产品信息向量化得到第三产品向量矩阵，然后将第三产品向量矩阵与第二词向量矩阵（客户的检索信息对应的向量矩阵）关联存储到指定的数据库中，以便后期作为训练样本使用。上述指定数据量是指存储上述第三产品向量矩阵与第二词向量矩阵的总的数据量。当数据量达到指定阈值，则说明样本数量达到要求，然后对上述的推荐模型进行训练，使用数据库中的样本数据继续对推荐模型进行训练，可以提高上述推荐模型的准确性。其训练方法与上述训练推荐模型的方法相同，在此不在赘述。上述对新客户购买的产品的信息进行向量化得第三产品向量矩阵的方法与上述将产品信息向量化的过程相同，在此同样不在赘述。
本申请实施例的产品销售的智能推荐装置，根据产品的产品信息的向量化矩阵，通过反向推荐模型计算出与其相对应的表示层向量，然后查找与该表示层向量相似度达到指定要求的第一词向量矩阵，而该第一词向量矩阵是利用客户的检索信息向量化得到的向量矩阵，所以可以根据第一词向量矩阵查找到对应客户的联系方式，因为检索信息是客户主动输入的信息，所以查找到的第一词向量矩阵对应的客户应该具有相对较高的购买欲望，以及相应的购买能力，所以将新的产品发送给对应的客户，推荐转销售的效率更高。
参照图3，本发明实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图3所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储反向推荐模型和推荐模型等。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现如上述各方法的实施例的流程。
本发明一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现如上述各方法的实施例的流程。
以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种产品销售的智能推荐方法，其特征在于，包括：
    将产品信息向量化，得到第一产品向量矩阵；
    将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
    到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
    提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
  2. 根据权利要求1所述的产品销售的智能推荐方法，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤之后，包括：
    获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
    若不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数；
    在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
  3. 根据权利要求1所述的产品销售的智能推荐方法，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤，包括：
    获取所述产品的售卖数据，以及购买所述产品的客户分布；
    将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户。
  4. 根据权利要求3所述的产品销售的智能推荐方法，其特征在于，所述将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户的步骤包括：
    将所述售卖数据和客户分布制作成可视化附图；
    将所述可视化附图以及所述产品的信息封装到一篇文档中形成所述推荐信息按照提取出的所述客户的联系方式推荐给客户。
  5. 根据权利要求1所述的产品销售的智能推荐方法，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤之后，包括：
    获取新客户的新检索信息，并将所述新检索信息向量化得到对应的第二词向量矩阵；
    将第二词向量矩阵输入到基于LSTM模型训练得到的推荐模型中学习，输出对应所述第二词向量矩阵的第二表示层向量；
    到训练好的对应产品的产品表示层向量矩阵中查找与所述第二表示层向量相似度最高的第三表示层向量；
    将第三表示层向量对应的产品输出，作为推荐给所述新客户的产品。
  6. 根据权利要求5所述的产品销售的智能推荐方法，其特征在于，所述将第三表示层向量对应的产品输出，作为推荐给所述新客户的产品的步骤之后，包括：
    记录新客户购买的产品，以及所述新客户的检索信息；
    将客户购买的产品的信息进行向量化得第三产品向量矩阵；
    将上所述新客户的新检索信息对应的第二词向量矩阵和所述第三产品向量矩阵关联地保存到指定的数据库中；
    当所述数据库中的数据量达到预设的阈值后，利用数据库中的全部第三产品向量矩阵和第二词向量矩阵对所述推荐模型进行继续训练，得到新的所述推荐模型。
  7. 根据权利要求5所述的产品销售的智能推荐方法，其特征在于，所述获取所述新客户的新检索信息，并将所述新检索信息向量化得到对应的所述第二词向量矩阵的步骤之前，包括：
    抓取公司网站上的全部产品的相关信息，形成各产品对应的产品信息；
    将各产品信息中重复出现的字提取出来，每一个重复的字只保留一个；
    将剩余的字输入到DSSM模型中计算，得出每一个字的词向量矩阵，以得到第一语料字典，所述第一语料字典用于向量化所述检索信息和所述产品信息。
  8. 一种产品销售的智能推荐装置，其特征在于，包括：
    第一向量化单元，用于将产品信息向量化，得到第一产品向量矩阵；
    反向学习单元，用于将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
    查找单元，用于到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
    提取推荐单元，用于提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
  9. 根据权利要求8所述的产品销售的智能推荐装置，其特征在于，还包括：
    获取判断单元，用于获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
    计数单元，用于若所述购买产品与所述产品不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数
    比较停用单元，用于在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
  10. 根据权利要求8所述的产品销售的智能推荐装置，其特征在于，所述提取推荐单元，包括：
    获取模块，用于获取所述产品的售卖数据，以及购买所述产品的客户分布；
    推荐模块，用于将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户。
  11. 根据权利要求10所述的产品销售的智能推荐装置，其特征在于，所述推荐模块，包括：
    可视化子单元，用于将所述售卖数据和客户分布制作成可视化附图；
    推荐子模块，用于将所述可视化附图以及所述产品的信息封装到一篇文档中形成所述推荐信息按照提取出的所述客户的联系方式推荐给客户。
  12. 根据权利要求8所述的产品销售的智能推荐装置，其特征在于，还包括：
    获取向量单元，用于获取新客户的新检索信息，并将所述新检索信息向量化得到对应的所述第二词向量矩阵；
    推荐学习单元，用于将第二词向量矩阵输入到基于LSTM模型训练得到的推荐模型中学习，输出对应所述第二词向量矩阵的第二表示层向量；
    像似查找单元，用于到训练好的对应产品的产品表示层向量矩阵中查找与所述第二表示层向量相似度最高的第三表示层向量；
    推荐单元，用于将第三表示层向量对应的产品输出，作为推荐给新客户的产品。
  13. 根据权利要求12所述的产品销售的智能推荐装置，其特征在于，还包括：
    记录单元，用于记录所述新客户购买的产品，以及所述新客户的检索信息；
    第二向量化单元，用于将所述新客户购买的产品的信息进行向量化得第三产品向量矩阵；
    关联单元，用于将上所述新客户的新检索信息对应的第二词向量矩阵和所述第三产品向量矩阵关联地保存到指定的数据库中；
    更新单元，用于当所述数据库中的数据量达到预设的阈值后，利用数据库中的全部第三产品向量矩阵和第二词向量矩阵对所述推荐模型进行继续训练，得到新的所述推荐模型。
  14. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现产品销售的智能推荐方法，该产品销售的智能推荐方法包括：
    将产品信息向量化，得到第一产品向量矩阵；
    将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
    到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
    提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
  15. 根据权利要求14所述的计算机设备，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤之后，包括：
    获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
    若不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数；
    在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
  16. 根据权利要求14所述的计算机设备，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤，包括：
    获取所述产品的售卖数据，以及购买所述产品的客户分布；
    将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户。
  17. 根据权利要求16所述的计算机设备，其特征在于，所述将所述产品的信息、售卖数据和客户分布形成推荐信息按照提取出的所述客户的联系方式推荐给客户的步骤包括：
    将所述售卖数据和客户分布制作成可视化附图；
    将所述可视化附图以及所述产品的信息封装到一篇文档中形成所述推荐信息按照提取出的所述客户的联系方式推荐给客户。
  18. 根据权利要求14所述的计算机设备，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤之后，包括：
    获取新客户的新检索信息，并将所述新检索信息向量化得到对应的第二词向量矩阵；
    将第二词向量矩阵输入到基于LSTM模型训练得到的推荐模型中学习，输出对应所述第二词向量矩阵的第二表示层向量；
    到训练好的对应产品的产品表示层向量矩阵中查找与所述第二表示层向量相似度最高的第三表示层向量；
    将第三表示层向量对应的产品输出，作为推荐给所述新客户的产品。
  19. 一种计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现产品销售的智能推荐方法，该产品销售的智能推荐方法包括：
    将产品信息向量化，得到第一产品向量矩阵；
    将所述第一产品向量矩阵输入到基于LSTM模型训练得到的反向推荐模型中，以输出对应所述第一产品向量矩阵的第一表示层向量；所述第一表示层向量是对应客户的检索信息的第一词向量矩阵的表示层向量；
    到预设的客户数据库中查找与所述第一表示层向量相似度达到指定要求的所述第一词向量矩阵，其中，所述第一词向量矩阵与所述客户的信息关联地存储在所述客户数据库中，所述客户的信息至少包括客户的联系方式；
    提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户。
  20. 根据权利要求14所述的计算机非易失性可读存储介质，其特征在于，所述提取所述第一词向量矩阵对应的客户的信息，并将所述产品信息按照提取出的所述客户的联系方式推荐给客户的步骤之后，包括：
    获取所述客户购买的购买产品，并判断所述购买产品与所述产品是否相同；
    若不相同，则在购买不同产品的计数的基础上加一，得到第一计数；若相同，则在购买相同产品的计数的基础上加一，得到的第二计数；
    在指定时间长度的时间节点处，使用第一计数比上第二计数，若比值大于预设阈值，则停用所述反向推荐模型。
PCT/CN2018/124392 2018-09-05 2019-01-21 产品销售的智能推荐方法、装置、计算机设备和存储介质 WO2020048062A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811032933.3A CN109389497A (zh) 2018-09-05 2018-09-05 产品销售的智能推荐方法、装置、计算机设备和存储介质
CN201811032933.3 2018-09-05

Publications (1)

Publication Number Publication Date
WO2020048062A1 true WO2020048062A1 (zh) 2020-03-12

Family

ID=65418471

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/124392 WO2020048062A1 (zh) 2018-09-05 2019-01-21 产品销售的智能推荐方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN109389497A (zh)
WO (1) WO2020048062A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113672816A (zh) * 2021-10-21 2021-11-19 腾讯科技（深圳）有限公司 帐号特征信息的生成方法、装置和存储介质及电子设备
CN113724053A (zh) * 2021-09-09 2021-11-30 内江师范学院 农产品供应链协调方法及系统
CN113743081A (zh) * 2021-09-03 2021-12-03 西安邮电大学 技术服务信息的推荐方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110069663B (zh) * 2019-04-29 2021-06-04 厦门美图之家科技有限公司 视频推荐方法及装置
CN113836379B (zh) * 2021-09-26 2023-08-25 北京百炼智能科技有限公司 一种基于客户画像的智能推荐方法和系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103514239A (zh) * 2012-11-26 2014-01-15 Tcl美国研究所 一种集成用户行为和物品内容的推荐方法及系统
CN103559623A (zh) * 2013-09-24 2014-02-05 浙江大学 一种基于联合非负矩阵分解的个性化产品推荐方法
CN103744966A (zh) * 2014-01-07 2014-04-23 Tcl集团股份有限公司 一种物品推荐方法、装置
US20170206581A1 (en) * 2016-01-15 2017-07-20 Target Brands, Inc. Product vector for product recommendation

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2554777A (en) * 2016-09-22 2018-04-11 Zensar Tech Limited A computer implemented interactive system and method for locating products and services

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103514239A (zh) * 2012-11-26 2014-01-15 Tcl美国研究所 一种集成用户行为和物品内容的推荐方法及系统
CN103559623A (zh) * 2013-09-24 2014-02-05 浙江大学 一种基于联合非负矩阵分解的个性化产品推荐方法
CN103744966A (zh) * 2014-01-07 2014-04-23 Tcl集团股份有限公司 一种物品推荐方法、装置
US20170206581A1 (en) * 2016-01-15 2017-07-20 Target Brands, Inc. Product vector for product recommendation

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113743081A (zh) * 2021-09-03 2021-12-03 西安邮电大学 技术服务信息的推荐方法
CN113743081B (zh) * 2021-09-03 2023-08-01 西安邮电大学 技术服务信息的推荐方法
CN113724053A (zh) * 2021-09-09 2021-11-30 内江师范学院 农产品供应链协调方法及系统
CN113724053B (zh) * 2021-09-09 2023-06-23 内江师范学院 农产品供应链协调方法及系统
CN113672816A (zh) * 2021-10-21 2021-11-19 腾讯科技（深圳）有限公司 帐号特征信息的生成方法、装置和存储介质及电子设备
CN113672816B (zh) * 2021-10-21 2022-02-08 腾讯科技（深圳）有限公司 帐号特征信息的生成方法、装置和存储介质及电子设备

Also Published As

Publication number Publication date
CN109389497A (zh) 2019-02-26

Similar Documents

Publication Publication Date Title
Athota et al. Chatbot for healthcare system using artificial intelligence
WO2020048062A1 (zh) 产品销售的智能推荐方法、装置、计算机设备和存储介质
WO2021000676A1 (zh) 问答方法、问答装置、计算机设备及存储介质
US11093560B2 (en) Stacked cross-modal matching
US7860347B2 (en) Image-based face search
US10496699B2 (en) Topic association and tagging for dense images
CN111538894B (zh) 查询反馈方法、装置、计算机设备及存储介质
US20190370273A1 (en) System, computer-implemented method and computer program product for information retrieval
JP6381775B2 (ja) 情報処理システム及び情報処理方法
WO2020048061A1 (zh) 产品推荐方法、装置、计算机设备和存储介质
CN111708873A (zh) 智能问答方法、装置、计算机设备和存储介质
CN109933785A (zh) 用于实体关联的方法、装置、设备和介质
CN110032728B (zh) 疾病名称标准化的转换方法和装置
US20220277005A1 (en) Semantic parsing of natural language query
WO2021196934A1 (zh) 一种基于字段相似度计算的问题推荐方法、装置和服务器
WO2020123689A1 (en) Suggesting text in an electronic document
CN111858940A (zh) 一种基于多头注意力的法律案例相似度计算方法及系统
Angadi et al. Multimodal sentiment analysis using reliefF feature selection and random forest classifier
CN113409907A (zh) 一种基于互联网医院的智能预问诊方法及系统
WO2022222942A1 (zh) 问答记录生成方法、装置、电子设备及存储介质
JP6446987B2 (ja) 映像選択装置、映像選択方法、映像選択プログラム、特徴量生成装置、特徴量生成方法及び特徴量生成プログラム
EP3901875A1 (en) Topic modelling of short medical inquiries
CN117435685A (zh) 文档检索方法、装置、计算机设备、存储介质和产品
CN115878761A (zh) 事件脉络生成方法、设备及介质
EP3731108A1 (en) Search system, search method, and program

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18932759

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18932759

Country of ref document: EP

Kind code of ref document: A1