New media anchor matching recommendation system and method based on deep learning and data analysis

Abstract

The invention relates to a new media anchor matching recommendation system and method based on deep learning and data analysis, which has the effect of improving the matching accuracy and effectiveness of a product and a new media anchor, and comprises the following steps: screening historical purchase orders about to-be-marketed product information from a commodity marketing database; the to-be-marketed product information comprises a product name; calculating a first proportion of the number of times that each feature tag appears in the historical purchase orders to the number of all the historical purchase orders; generating a key feature tag set by using the feature tags of which the first proportions exceed a preset proportion; distributing an initial weight value for each feature tag in the key feature tag set based on the size of the first proportion of each feature tag; generating a real weight value of the reference feature tag; calculating the matching degree of the concerned user information of each new media anchor based on the real weight value of each feature tag and the concerned user information of each new media anchor in a new media anchor database; and sorting and generating a recommended new media anchor sequence according to the matching degree.

Description

Technical Field

[0001] The present invention relates to the technical field of network marketing, and in particular to a new media anchor matching recommendation system and method based on deep learning and data analysis. Background Art

[0002] With the rapid development of Internet technology, the e-commerce industry has ushered in unprecedented prosperity. In this era of information explosion, how to effectively promote products to target users has become a key issue that e-commerce companies need to solve urgently. New media anchor marketing, as an emerging marketing method, has gradually become the preferred strategy for many brands and product promotions with its powerful influence and precise audience positioning.

[0003] However, the success of new media anchor marketing depends largely on the matching degree between the new media anchor and the product to be marketed. Traditional matching methods often rely on manual judgment or simple data analysis, which is not only inefficient but also difficult to guarantee the accuracy and effectiveness of the matching results. Therefore, how to develop an efficient and accurate new media anchor matching recommendation system has become a technical problem that the e-commerce industry needs to solve urgently. Summary of the invention

[0004] In order to improve the accuracy and effectiveness of matching products and new media anchors, the present application provides a new media anchor matching recommendation system and method based on deep learning and data analysis.

[0005] The above-mentioned invention objective of the present application is achieved through the following technical solutions: A new media anchor matching recommendation method based on deep learning and data analysis, comprising: In response to the product information to be marketed, historical purchase orders related to the product information to be marketed are screened in a commodity marketing database, wherein the historical purchase orders include an order code and multiple characteristic tags of the ordering user; the product information to be marketed includes a product name; Calculate the first proportion of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders; Generate a key feature tag set from the feature tags whose first proportion exceeds the preset proportion; Based on the size of the first proportion of each feature tag, an initial weight value is assigned to each feature tag in the key feature tag set; A certain feature tag in the key feature tag set is used as a benchmark feature tag, and other feature tags are used as comparison feature tags. The number of historical purchase orders coexisting between the benchmark feature tag and another comparison feature tag is compared in pairs. Based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, the real weight value of the benchmark feature tag is generated, and the real weight value of each feature tag in the key feature tag set is calculated in sequence; Calculate the matching degree of each new media anchor's attention user information based on the real weight value of each feature tag and the attention user information of each new media anchor in the new media anchor database; the attention user information includes the attention user ID and multiple feature tags of the attention user;

[0006] Sort by matching degree and generate a recommended new media anchor sequence.

[0007] In a preferred example, the present application can be further configured as follows: based on the real weight value of each feature tag and the follower information of each new media anchor in the new media anchor database, the matching degree of the follower information of each new media anchor is calculated, including: Generate multiple key feature label subsets based on the key feature label set, different key feature label subsets contain elements in different key feature label sets, contain different numbers of elements in the key feature label sets as containing elements in different key feature label sets, contain the same number of elements in different key feature label sets as containing elements in different key feature label sets; Count the number of key feature tag subsets included in the follow-up user information of the new media anchor to obtain the number of appearances of each key feature tag subset; The matching degree of the new media anchor's follower information is calculated based on the number of occurrences of each key feature tag subset and the true weight value of each feature tag.

[0008] In a preferred example, the present application can be further configured as follows: based on the number of occurrences of each key feature tag subset and the real weight value of each feature tag, the matching degree of the new media anchor's attention user information is calculated, including: Calculate the ratio of the number of appearances of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance ratio of each key feature tag subset; The sub-matching degree of the key feature tag subset is obtained by calculating the sum of the products of the real weight value of each feature tag in the key feature tag subset and the emergence ratio; The sum of the sub-matching degrees of each key feature tag subset is used as the matching degree of the new media anchor's follow-up user information.

[0009] In a preferred example, the present application may be further configured to generate a real weight value of the benchmark feature tag based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, including: Calculated by the following formula: , Among them, t is the number of times the benchmark feature label appears in the historical purchase orders, s is the number of co-existing historical purchase orders, q is the initial weight value of the benchmark feature label, and Q is the true weight value of the benchmark feature label.

[0010] In a preferred example, the present application can be further configured as follows: The commission price and the number of followed user information of each new media anchor are obtained, and the information is associated with the new media anchor sequence.

[0011] The second object of the invention is achieved by the following technical solutions: A system based on any of the above-mentioned new media anchor matching recommendation methods based on deep learning and data analysis, comprising: a response screening module, for responding to the information of the product to be marketed and screening out historical purchase orders related to the information of the product to be marketed in the commodity marketing database, wherein the historical purchase orders include order codes and multiple characteristic tags of ordering users; the information of the product to be marketed includes product names; A ratio calculation module, used to calculate the first ratio of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders; A set generation module, used for generating a key feature tag set from feature tags whose first proportion exceeds a preset proportion; A weight allocation module, used to allocate an initial weight value to each feature tag based on the size of the first proportion of each feature tag; A weight calculation module is used to use a certain feature tag in the key feature tag set as a benchmark feature tag and other feature tags as comparison feature tags, and traverse and compare the number of historical purchase orders coexisting between the benchmark feature tag and another comparison feature tag, and generate a real weight value of the benchmark feature tag based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, and sequentially calculate the real weight value of each feature tag in the key feature tag set; A matching degree calculation module is used to calculate the matching degree of each new media anchor's followed user information based on the real weight value of each feature tag and the followed user information of each new media anchor in the new media anchor database; the followed user information includes the followed user ID and multiple feature tags of the followed user; The recommendation module is used to sort and generate a sequence of recommended new media anchors according to the matching degree.

[0012] In a preferred example, the present application can be further configured to include: A subset generating unit, configured to generate a plurality of key feature label subsets based on the key feature label set, wherein different key feature label subsets contain elements in different key feature label sets, elements in different numbers of key feature label sets are deemed to contain elements in different key feature label sets, and elements in the same number of key feature label sets but in different key feature label sets are deemed to contain elements in different key feature label sets; A quantity counting unit is used to count the number of key feature tag subsets included in the follower information of the new media anchor to obtain the number of appearances of each key feature tag subset; if more are included, it is counted once, and if less are included, it is not counted; The matching degree calculation unit is used to calculate the matching degree of the new media anchor's follow-up user information based on the number of appearances of each key feature tag subset and the real weight value of each feature tag.

[0013] In a preferred example, the present application can be further configured to include: The proportion calculation subunit is used to calculate the proportion of the appearance times of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance proportion of each key feature tag subset; The sub-matching degree calculation sub-unit is used to calculate the sum of the products of the real weight value of each feature tag in the key feature tag subset and the emergence ratio of the key feature tag subset to obtain the sub-matching degree of the key feature tag subset; and a calculation subunit, which is used to take the sum of the sub-matching degrees of each key feature tag subset as the matching degree of the new media anchor's attention user information.

[0014] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application can more carefully describe the characteristics of the user groups that new media anchors follow through multiple subsets of key feature tags. Each subset contains different combinations of feature tags, which can more comprehensively reflect the diversity of user groups. And by combining the real weight values ​​of each feature tag for calculation, it can ensure that when calculating the matching degree, more important feature tags have a greater impact on the results, thereby improving the accuracy of the matching. It can achieve accurate mining and efficient matching of the deep-level correlation between new media anchors and the products to be marketed; 2. The real weight value mainly adjusts the weight of each feature tag in the key feature tag set. Based on this formula, if the number of historical purchase orders that a feature tag coexists with other feature tags is less than the number of times the feature tag itself appears in historical purchase orders, the weight of the feature tag will be increased significantly. If the number of historical purchase orders that a feature tag coexists with other feature tags is greater than the number of times the feature tag itself appears in historical purchase orders, the weight of the feature tag will be increased slightly. If a feature tag itself has a large first proportion, the real weight value will also be increased significantly. If a feature tag itself has a small first proportion, but a small number of coexisting historical purchase orders, and compared with other feature tags, the number of times the feature tag appears alone in historical purchase orders accounts for a large proportion, then the real weight value of the feature tag will also be relatively increased. The weight of a feature tag with a small first proportion but a large proportion of appearances in historical purchase orders can be increased, which is more in line with the actual situation and more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flowchart of a new media anchor matching recommendation method based on deep learning and data analysis in one embodiment of the present application;

[0016] Figure 2 This is a module connection diagram of a new media anchor matching recommendation system based on deep learning and data analysis in one embodiment of the present application. DETAILED DESCRIPTION

[0017] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0018] It should be noted that the terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure.

[0019] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.

[0020] In the prior art, although there are some recommendation systems based on data analysis, most of them are limited to simple statistics and analysis of user behavior data, and lack the exploration of the deep correlation between new media anchors and products. For example, some systems only recommend related products or new media anchors based on the user's purchase history or browsing history, but ignore the potential connection between the user's feature tags and the feature tags of the users that the new media anchors follow.

[0021] Figure 1 is a flowchart of a new media anchor matching recommendation method based on deep learning and data analysis in an embodiment of the present application. Figure 1 As shown, the new media anchor matching recommendation method based on deep learning and data analysis includes: S1. Responding to the information of the product to be marketed and filtering out historical purchase orders related to the information of the product to be marketed in a commodity marketing database.

[0022] Among them, the information of the products to be marketed includes the product name; the historical purchase orders include the order code and multiple characteristic tags of the ordering user; the construction of the commodity marketing database is through the collection of previous commodity marketing order data, such as order code, ordering user ID, order time, order amount, purchase product details, etc., the type and quantity of characteristic tags of the ordering user, these tags may include the age range, gender, region, consumption habits, hobbies, aesthetic preferences, etc. of the ordering user. The acquisition of characteristic tags can be through the analysis of the user's purchase history, customer service record analysis, user survey and questionnaire analysis, account behavior analysis, browsing history and click behavior analysis, etc., and can also be analyzed in a variety of ways, and the analysis method can be to use a neural network model to infer the input data. Specifically, for example, collect user behavior data on social media, browsing history and click behavior data and other interest-related data; clean the collected data to remove noise, outliers and duplicate data; extract features that can reflect user consumption habits, interests and aesthetic preferences from the preprocessed data; convert the extracted features into a form suitable for neural network input, such as numerical features, categorical features (represented by unique hot encoding or embedded vectors), etc.; select a suitable neural network architecture according to data characteristics and task requirements, such as multi-layer perceptron (MLP) or convolutional neural network (CNN), etc.; set the input layer to receive feature data, and for categorical features, set the embedding layer to convert them into low-dimensional vectors. Then build multiple hidden layers, capture the complex relationship between features through nonlinear activation functions (such as ReLU, sigmoid, etc.); design the output layer as multiple independent binary classification tasks (i.e., multi-label classification), each label corresponds to an output node, and use the sigmoid function to output the probability value of each label, so that the model can predict multiple labels at the same time. Among them, the label can be a predefined category such as consumption habits, hobbies, aesthetic preferences, etc. The binary cross-entropy loss function is selected as the loss function of the model, which is suitable for multi-label classification tasks. For each label, a binary cross-entropy loss is calculated, and then the losses of all labels are added to get the total loss; the weights and biases of the neural network are updated using the gradient descent algorithm or its variants (such as Adam, RMSprop, etc.). The performance of the model is optimized by adjusting hyperparameters such as learning rate, batch size, number of neurons in the hidden layer, regularization method (such as L1, L2 regularization); in addition, hyperparameter tuning can be performed using methods such as grid search, random search or Bayesian optimization. After the model training is completed, the user's features are input into the model, and the output layer will give the probability distribution of each predefined label. According to the set threshold (such as 0.5), the label with a probability greater than the threshold is used as the user's feature label; thus, each user can get a set containing multiple feature labels.

[0023] S2. Calculate the first proportion of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders.

[0024] Specifically, a first proportion of the number of times each feature tag appears in the aforementioned filtered historical purchase orders to the number of all the aforementioned filtered historical purchase orders is calculated.

[0025] S3. Generate a key feature tag set from the feature tags whose first proportion exceeds the preset proportion.

[0026] The preset proportion may be a value such as 20%, 10%, etc. If it is 10%, that is, 10% of all historical purchase orders have this feature tag, then this feature tag belongs to the key feature tag set.

[0027] S4. Allocate an initial weight value to each feature tag in the key feature tag set based on the size of the first proportion of each feature tag.

[0028] The ratio of the first proportion value of each feature tag to the sum of the first proportions of each feature tag in the key feature tag set is used as the initial weight value of the feature tag.

[0029] For example, in the key feature label set {a, b, c, d, e, f, g}, abcdefg are 7 feature labels, among which the first proportion of a is 70%, the first proportion of b is 60%, the first proportion of c is 53%, the first proportion of d is 40%, the first proportion of e is 31%, the first proportion of f is 30%, and the first proportion of g is 23%. Then the initial weight value of a is 0.228, the initial weight value of b is 0.195, and so on.

[0030] S5. Take a certain feature tag in the key feature tag set as a baseline feature tag and other feature tags as comparison feature tags. Compare the number of coexisting historical purchase orders between the baseline feature tag and another comparison feature tag. Generate the true weight value of the baseline feature tag based on the number of coexisting historical purchase orders, the number of times the baseline feature tag appears in the historical purchase orders and the initial weight value of the baseline feature tag. Calculate the true weight value of each feature tag in the key feature tag set in turn.

[0031] For example, a feature label a is used as the baseline feature label, and the remaining six feature labels bcdefg are comparison feature labels. The pairwise traversal comparison refers to comparing the baseline feature label with the remaining six feature labels one by one, that is, after a is compared with b, a is compared with c again, until a is compared with g, to achieve traversal comparison. The specific comparison is to obtain the number of coexisting and historical purchase orders between the baseline feature label and a comparison feature label, that is, the number of times they appear in the historical purchase orders at the same time. Based on the number of coexisting historical purchase orders, the number of times the baseline feature label appears in the historical purchase orders, and the initial weight value of the baseline feature label, the real weight value of the baseline feature label is generated, including: Calculated by the following formula: , Among them, t is the number of times the benchmark feature label appears in the historical purchase orders, n is the number of feature labels in the key feature label set, s is the sum of the number of co-existence and historical purchase orders between the benchmark feature label and the remaining comparison feature labels, q is the initial weight value of the benchmark feature label, and Q is the true weight value of the benchmark feature label.

[0032] The real weight value mainly adjusts the weight of each feature tag in the key feature tag set. It can be understood that based on this formula, if the number of historical purchase orders that a feature tag coexists with other feature tags is less than the number of times the feature tag itself appears in historical purchase orders, then the weight of the feature tag will increase significantly. If the number of historical purchase orders that a feature tag coexists with other feature tags is greater than the number of times the feature tag itself appears in historical purchase orders, then the weight of the feature tag will increase slightly. If a feature tag itself has a large first proportion, then the real weight value will also increase significantly. If a feature tag itself has a small first proportion, but the number of coexisting historical purchase orders is small, and compared with other feature tags, the number of times the feature tag appears alone in historical purchase orders accounts for a large proportion, then the real weight value of the feature tag will also increase relatively significantly.

[0033] S6. Calculate the matching degree of the user information of each new media anchor based on the real weight value of each feature tag and the user information of each new media anchor in the new media anchor database.

[0034] The followed user information includes the followed user ID and multiple characteristic tags of the followed user; the characteristic tags are also obtained by analyzing in the above-mentioned manner.

[0035] S6 includes: S61. Generate multiple key feature tag subsets based on the key feature tag set.

[0036] Among them, different key feature label subsets contain elements from different key feature label sets, elements containing different numbers of key feature label sets are regarded as containing elements from different key feature label sets, and elements containing the same number but different key feature label sets are regarded as containing elements from different key feature label sets; that is, multiple key feature label subsets are obtained by traversing the combination methods of each element in the key feature label set.

[0037] S62, counting the number of key feature tag subsets included in the follow-up user information of the new media anchor to obtain the appearance times of each key feature tag subset.

[0038] It is understandable that, in this step, if all elements are included in the key feature tag subset of another key feature tag subset, only the number of occurrences of the key feature tag subset with the largest number of elements is counted. That is, if a certain focused user information contains the key feature tag subset {a,b,c,e}, only the key feature tag subset {a,b,c,e} is counted once, and the key feature tag subsets {a,b,c}, {b,c,e}{a,e}, etc., which have fewer elements than the key feature tag subset {a,b,c,e} and all elements are included in the key feature tag subset {a,b,c,e}, are not counted.

[0039] S63, calculating the matching degree of the new media anchor's follower information based on the appearance times of each key feature tag subset and the real weight value of each feature tag.

[0040] Specifically, S63 includes: S631. Calculate the ratio of the number of appearances of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance ratio of each key feature tag subset.

[0041] S632: Calculate the sum of the products of the real weight value and the emergence proportion of each feature tag in the key feature tag subset to obtain the sub-matching degree of the key feature tag subset.

[0042] For example, the sub-matching degree of the key feature label subset {a, b, c, e} is Qa*Fa+Qb*Fa+Qc*Fa+Qe*Fa, where F is the emergence ratio.

[0043] S633: The sum of the sub-matching degrees of each key feature tag subset is used as the matching degree of the new media anchor's attention user information.

[0044] S7. Sort by matching degree and generate a recommended new media anchor sequence.

[0045] In one embodiment, it further includes: S8. Obtain the commission price and the number of following users of each new media anchor in the new media anchor database, and associate them in the new media anchor sequence. This allows the selection personnel to view the commission price, the number of fans and the matching degree at a glance, so as to better select the corresponding new media anchor for promotion according to the budget.

[0046] This application fully considers the weight differences of feature tags. In practical applications, different feature tags have different importance for describing the characteristics of users or new media anchors. Therefore, simply considering all feature tags as equivalent and making recommendations based on them will undoubtedly reduce the accuracy and pertinence of the recommendations. This application can more carefully depict the characteristics of the user groups that new media anchors follow through multiple subsets of key feature tags. Each subset contains different combinations of feature tags, which can more comprehensively reflect the diversity of user groups. And by combining the true weight values ​​of each feature tag for calculation, it can be ensured that when calculating the matching degree, more important feature tags have a greater impact on the results, thereby improving the accuracy of the matching. It can achieve accurate mining and efficient matching of the deep correlation between new media anchors and the products to be marketed.

[0047] The present application also provides a new media anchor matching recommendation system based on deep learning and data analysis, which applies the aforementioned new media anchor matching recommendation method based on deep learning and data analysis, referring to Figure 2 ,include: The response screening module is used to respond to the information of the products to be marketed and screen out historical purchase orders related to the information of the products to be marketed in the commodity marketing database, where the historical purchase orders include order codes and multiple feature tags of the ordering users; the information of the products to be marketed includes product names.

[0048] The proportion calculation module is used to calculate the first proportion of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders.

[0049] A set generation module, used for generating a key feature tag set from feature tags whose first proportion exceeds a preset proportion; The weight allocation module is used to allocate an initial weight value to each feature tag based on the size of the first proportion of each feature tag.

[0050] The weight calculation module is used to take a certain feature tag in the key feature tag set as a baseline feature tag and other feature tags as comparison feature tags, and traverse and compare the number of co-existing historical purchase orders between the baseline feature tag and another comparison feature tag. Based on the number of co-existing historical purchase orders, the number of times the baseline feature tag appears in the historical purchase orders and the initial weight value of the baseline feature tag, the true weight value of the baseline feature tag is generated, and the true weight value of each feature tag in the key feature tag set is calculated in turn.

[0051] The matching degree calculation module is used to calculate the matching degree of each new media anchor's followed user information based on the real weight value of each feature tag and the followed user information of each new media anchor in the new media anchor database; the followed user information includes the followed user ID and multiple feature tags of the followed user.

[0052] The recommendation module is used to sort and generate a sequence of recommended new media anchors according to the matching degree.

[0053] In one embodiment, the matching degree calculation module includes: A subset generation unit is used to generate multiple key feature label subsets based on the key feature label set, different key feature label subsets contain elements in different key feature label sets, elements contained in different numbers of key feature label sets are regarded as containing elements in different key feature label sets, and elements contained in the same number but different key feature label sets are regarded as containing elements in different key feature label sets.

[0054] The quantity counting unit is used to count the number of key feature tag subsets included in the follow-up user information of the new media anchor to obtain the number of appearances of each key feature tag subset.

[0055] The matching degree calculation unit is used to calculate the matching degree of the new media anchor's follow-up user information based on the number of appearances of each key feature tag subset and the real weight value of each feature tag.

[0056] Furthermore, the matching degree calculation unit includes: The proportion calculation subunit is used to calculate the proportion of the appearance times of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance proportion of each key feature tag subset; The sub-matching degree calculation sub-unit is used to calculate the sum of the products of the real weight value of each feature tag in the key feature tag subset and the emergence ratio of the key feature tag subset to obtain the sub-matching degree of the key feature tag subset; and a calculation subunit, which is used to take the sum of the sub-matching degrees of each key feature tag subset as the matching degree of the new media anchor's attention user information.

[0057] For the specific limitations of the new media anchor matching recommendation system based on deep learning and data analysis, please refer to the limitations of the new media anchor matching recommendation method based on deep learning and data analysis above, which will not be repeated here. Each step of the above-mentioned new media anchor matching recommendation method based on deep learning and data analysis can be implemented in whole or in part by software, hardware and a combination thereof.

[0058] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0059] These computer programs (also referred to as programs, software, software applications, or code) include machine instructions for programmable processors and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or means (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0060] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0061] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0062] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0063] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A new media anchor matching recommendation method based on deep learning and data analysis, characterized in that: include: In response to the product information to be marketed, historical purchase orders related to the product information to be marketed are screened in a commodity marketing database, wherein the historical purchase orders include an order code and multiple characteristic tags of the ordering user; the product information to be marketed includes a product name; Calculate the first proportion of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders; Generate a key feature tag set from the feature tags whose first proportion exceeds the preset proportion; Based on the size of the first proportion of each feature tag, an initial weight value is assigned to each feature tag in the key feature tag set; A certain feature tag in the key feature tag set is used as a benchmark feature tag, and other feature tags are used as comparison feature tags. The number of historical purchase orders coexisting between the benchmark feature tag and another comparison feature tag is compared in pairs. Based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, the real weight value of the benchmark feature tag is generated, and the real weight value of each feature tag in the key feature tag set is calculated in sequence; Calculate the matching degree of each new media anchor's attention user information based on the real weight value of each feature tag and the attention user information of each new media anchor in the new media anchor database; the attention user information includes the attention user ID and multiple feature tags of the attention user; Sort by matching degree and generate a recommended new media anchor sequence.

2. The new media anchor matching recommendation method based on deep learning and data analysis as claimed in claim 1, characterized in that: The matching degree of the following user information of each new media anchor is calculated based on the real weight value of each feature tag and the following user information of each new media anchor in the new media anchor database, including: Generate multiple key feature label subsets based on the key feature label set, different key feature label subsets contain elements in different key feature label sets, contain different numbers of elements in the key feature label sets as containing elements in different key feature label sets, contain the same number of elements in different key feature label sets as containing elements in different key feature label sets; Count the number of key feature tag subsets included in the follow-up user information of the new media anchor to obtain the number of appearances of each key feature tag subset; The matching degree of the new media anchor's follower information is calculated based on the number of occurrences of each key feature tag subset and the true weight value of each feature tag.

3. The new media anchor matching recommendation method based on deep learning and data analysis as claimed in claim 2, characterized in that: The matching degree of the new media anchor’s follower information is calculated based on the number of occurrences of each key feature tag subset and the real weight value of each feature tag, including: Calculate the ratio of the number of appearances of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance ratio of each key feature tag subset; The sub-matching degree of the key feature tag subset is obtained by calculating the sum of the products of the true weight value of each feature tag in the key feature tag subset and the emergence ratio; The sum of the sub-matching degrees of each key feature tag subset is taken as the matching degree of the new media anchor's follow-up user information.

4. The new media anchor matching recommendation method based on deep learning and data analysis as claimed in claim 1, characterized in that: Based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, a real weight value of the benchmark feature tag is generated, including: Calculated by the following formula: ， Among them, t is the number of times the benchmark feature label appears in the historical purchase orders, s is the number of co-existing historical purchase orders, q is the initial weight value of the benchmark feature label, and Q is the true weight value of the benchmark feature label.

5. The new media anchor matching recommendation method based on deep learning and data analysis as claimed in claim 1, characterized in that: Also includes: The commission price and the number of followed user information of each new media anchor are obtained, and the information is associated with the new media anchor sequence.

6. A new media anchor matching recommendation system based on deep learning and data analysis, characterized in that: Executing the new media anchor matching recommendation method based on deep learning and data analysis as described in any one of claims 1 to 5, comprising: a response screening module, for responding to the information of the product to be marketed and screening out historical purchase orders related to the information of the product to be marketed in the commodity marketing database, wherein the historical purchase orders include order codes and multiple characteristic tags of ordering users; the information of the product to be marketed includes product names; A ratio calculation module, used to calculate the first ratio of the number of times each feature tag appears in the historical purchase orders to the number of all historical purchase orders; A set generation module, used for generating a key feature tag set from feature tags whose first proportion exceeds a preset proportion; A weight allocation module, used to allocate an initial weight value to each feature tag based on the size of the first proportion of each feature tag; A weight calculation module is used to use a certain feature tag in the key feature tag set as a benchmark feature tag and other feature tags as comparison feature tags, and traverse and compare the number of historical purchase orders coexisting between the benchmark feature tag and another comparison feature tag, and generate a real weight value of the benchmark feature tag based on the number of coexisting historical purchase orders, the number of times the benchmark feature tag appears in the historical purchase orders, and the initial weight value of the benchmark feature tag, and sequentially calculate the real weight value of each feature tag in the key feature tag set; A matching degree calculation module is used to calculate the matching degree of each new media anchor's followed user information based on the real weight value of each feature tag and the followed user information of each new media anchor in the new media anchor database; the followed user information includes the followed user ID and multiple feature tags of the followed user; The recommendation module is used to sort and generate a sequence of recommended new media anchors according to the matching degree.

7. The new media anchor matching recommendation system based on deep learning and data analysis as claimed in claim 6, characterized in that: The matching degree calculation module includes: A subset generation unit is used to generate multiple key feature label subsets based on the key feature label set, different key feature label subsets contain elements from different key feature label sets, elements contained in different numbers of key feature label sets are deemed to contain elements from different key feature label sets, and elements contained in the same number of key feature label sets but in different elements are deemed to contain elements from different key feature label sets.

8. The new media anchor matching recommendation system based on deep learning and data analysis as claimed in claim 6, characterized in that: The matching degree calculation module includes: The quantity counting unit is used to count the number of key feature tag subsets included in the follow-up user information of the new media anchor to obtain the number of appearances of each key feature tag subset.

9. The new media anchor matching recommendation system based on deep learning and data analysis as claimed in claim 6 or 7, characterized in that: The matching degree calculation module includes: The matching degree calculation unit is used to calculate the matching degree of the new media anchor's follow-up user information based on the number of appearances of each key feature tag subset and the real weight value of each feature tag.

10. The new media anchor matching recommendation system based on deep learning and data analysis as claimed in claim 9, characterized in that: The matching degree calculation unit includes: The proportion calculation subunit is used to calculate the proportion of the appearance times of each key feature tag subset to the total number of user information followed by the new media anchor to obtain the appearance proportion of each key feature tag subset; The sub-matching degree calculation sub-unit is used to calculate the sum of the products of the real weight value of each feature tag in the key feature tag subset and the emergence ratio of the key feature tag subset to obtain the sub-matching degree of the key feature tag subset; and a calculation subunit, which is used to take the sum of the sub-matching degrees of each key feature tag subset as the matching degree of the new media anchor's attention user information.

Images