Sales auxiliary system and method based on AI technology

Abstract

The invention discloses a sales assistance system and method based on an AI technology, and relates to the technical field of sales assistance, and the system comprises a product data collection unit, and is applied to a live broadcast carrier to sell products for a product manufacturer in a commission mode. The product data collection unit is used for acquiring historical sales data of products which need to be sold by a product manufacturer in a live broadcast manner, processing the acquired historical sales data of the products, and storing the processed historical sales data of the products. The invention provides a sales auxiliary system and method based on the AI technology, and the system is used for a live-broadcasting carrier to sell a product for a product manufacturer in a commission mode, and obtains the historical sales data of the product through a set product data collection unit. And local topological feature extraction and vectorization processing are carried out on the obtained historical sales data of the product.

Description

Technical Field

[0001] The present invention relates to the field of sales assistance technology, and specifically to a sales assistance system and method based on AI technology. Background Art

[0002] In the process of realizing social informatization and digitalization through the Internet, the retail industry has developed rapidly relying on the development and changes of science and technology, and at the same time has brought more challenges. In the Chinese patent with application number 201810391849.4, "a commodity sales assistance system based on visual recognition technology, belonging to the field of commodity sales assistance, solves the problem of excessive reliance on shopping guides in the existing offline retail industry, commodity sales assistance system: the effective movement path generation module generates the effective movement path of the customer in the identification area, the entry sequence acquisition module and the exit sequence acquisition module respectively obtain the order of the customer entering and exiting the effective sub-identification area, the distance acquisition module obtains the distance from each effective sub-identification area to the starting point of the effective movement path, the stay time acquisition module determines the customer's stay time in each effective sub-identification area, the repeated passing detection module determines the effective sub-identification area that is repeatedly passed and the number of times the area is repeatedly passed, the candidate intention identification area judgment module judges a predetermined number of candidate intention identification areas based on the above parameters, and uses them as input parameters of the intention identification area judgment module, and the commodity information display module displays the information of the intended commodity according to the output result of the intention identification area judgment module."; This existing technology only solves a major drawback of the existing offline retail industry, which is that its business model is overly dependent on shopping guides. After customers enter the store and select the intended products, they often need shopping guides to provide them with more product information in a timely manner. At this time, if the shopping guide has no time to take care of the customer, cannot accompany the customer to select products all the time, or does not have complete information about the products, it is likely to lead to the loss of the customer and the failure of the transaction. It does not take into account that when selling the products of product manufacturers through live streaming, the product sales data during the historical live streaming should be processed, so that the processing results can be used as supplementary features for subsequent model training, so as to improve the prediction accuracy of product sales data, and based on the prediction results of product sales data, the marketing costs of live streaming merchants are decided on a case-by-case basis. Summary of the invention

[0003] The purpose of the present invention is to provide a sales assistance system and method based on AI technology to solve the problems raised in the above background technology.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a sales assistance system based on AI technology, including a product data collection unit, the sales assistance system based on AI technology is applied to a live broadcasting merchant to sell products for a product manufacturer under a commission model, the product data collection unit acquires the historical sales data of the products that the product manufacturer needs the live broadcasting merchant to sell, processes the acquired product historical sales data, and stores the processed product historical sales data; A product sales forecasting unit, which aims to forecast product sales volume, constructs a product sales forecasting model and a model training data set, trains the product sales forecasting model and verifies its accuracy through the model training data set, and obtains product sales volume data at the forecasting moment selected by the live streaming merchant through the trained product sales forecasting model; A sales strategy formulation unit, which builds a live streaming sales supply chain based on product manufacturers, live streaming sales providers, and consumers, and formulates live streaming sales strategies based on different situations based on product sales volume data at a predicted time selected by a user; The product sales assistance unit is provided with a prediction backtracking function, which modifies the original data input into the product sales prediction model, compares the prediction result with the original prediction result, thereby determining the impact on the original prediction result, and further realizing product sales assistance.

[0005] Preferably, the product data collection unit includes a historical sales data collection module and a historical sales data processing module. The historical sales data collection module acquires the historical sales data of the products that the product manufacturer needs the live streaming sales merchant to sell, including the sales data during the historical live streaming and the self-sales data of the network platform. The historical sales data processing module is electrically connected to the historical sales data collection module. The historical sales data processing module processes the acquired product historical sales data. The specific operation is to extract topological features from the product historical time series sales data without a topological spatial structure through the Witness complex construction method, and in order to make the extracted topological features contain the data connection between the previous and next times, a local topological feature is calculated based on the product historical time series sales data on a sliding window of a size of n time series by extracting local topological features, so that each local topological feature contains the information content of the evolution of the local topological feature in the time series historical sales data of size n over time, and the local topological feature is combined with the actual value in the corresponding product historical time series sales data through a persistent homology method based on a sliding window.

[0006] Preferably, the product data collection unit also includes a historical sales data vectorization module and a historical sales data storage module, the historical sales data vectorization module and the historical sales data processing module are electrically connected, the historical sales data vectorization module converts local topological features into vector form by utilizing the method of vectorizing the persistence graph using kernel embedding, so that it can be subsequently input into the product sales prediction model for use, the historical sales data storage module and the historical sales data vectorization module are electrically connected, the historical sales data storage module uploads the local topological features converted into vector form to the system's built-in database for storage, and synchronously uploads the product historical time series sales data combined with the corresponding local topological features to the system's built-in database for storage.

[0007] Preferably, the product sales forecasting unit includes a product sales forecasting model construction module and a model training data set construction module. The product sales forecasting model construction module aims at predicting product sales volume and constructs a product sales forecasting model based on the Nbeats model. A multi-head attention mechanism is introduced on the basis of the original Nbeats model, so that the product sales forecasting model pays attention to the local topological features extracted from the product historical time series sales data. At the same time, the product sales forecasting model learns the topological structure contained in the product historical time series sales data, and uses the local topological features extracted from the product historical time series sales data as supplementary features for subsequent model training, thereby improving the prediction accuracy of the product sales forecasting model. The model training data set construction module crawls product sales data samples through a web crawler, extracts local topological features and vectorizes the crawled product sales data samples, and constructs a model training data set based on the processed product sales data samples.

[0008] Preferably, the product sales forecasting unit further includes a product sales forecasting model training module and a product sales forecasting model output module, the product sales forecasting model training module is electrically connected to the product sales forecasting model construction module and the model training data set construction module respectively, the product sales forecasting model training module performs model training on the product sales forecasting model through the model training data set, and performs accuracy verification operation on the trained product sales forecasting model through the model accuracy algorithm, and the specific calculation formula is as follows:

[0009] In the formula, represents the accuracy value of the product sales forecast model, Indicates the number of data samples predicted correctly, The product sales prediction model output module represents the number of data samples with prediction errors. The product sales prediction model output module is electrically connected to the product sales prediction model training module. The product sales prediction model output module uses the local topological features in the form of vectors corresponding to the historical product sales data during live streaming stored in the database as the input of the product sales prediction model, and obtains the product sales volume data output by the product sales prediction model at the prediction time selected by the live streaming merchant.

[0010] Preferably, the sales strategy formulation unit includes a live streaming e-commerce supply chain construction module, which constructs a live streaming e-commerce supply chain based on product manufacturers, live streaming e-commerce merchants and consumers, and between product manufacturers and live streaming e-commerce merchants, product manufacturers need to pay product sales commissions to live streaming e-commerce merchants, between product manufacturers and consumers, product manufacturers need to bear product quality costs, and between live streaming e-commerce merchants and consumers, live streaming e-commerce merchants need to bear live streaming marketing costs.

[0011] Preferably, the sales strategy formulation unit also includes a situation-based sales strategy formulation module, which is electrically connected to the live streaming supply chain construction module. The situation-based sales strategy formulation module outputs product sales volume data based on the product sales forecasting model at the forecast time selected by the live streaming merchant, and compares it with the product sales threshold set by the live streaming merchant. When the product sales volume data output by the product sales forecasting model at the forecast time selected by the live streaming merchant is greater than the product sales threshold set by the live streaming merchant, the live streaming merchant needs to formulate a live streaming sales strategy while jointly bearing the product quality cost with the product manufacturer. The specific operation is that the live streaming merchant first determines the proportion of the product quality cost to be borne, and the product manufacturer then determines the sales price and product quality cost of the product. The live streaming merchant finally decides the live streaming marketing cost based on the sales price and product quality cost of the product. First, a profit function 1 of the live streaming merchant is constructed, and its specific expression is as follows:

[0012] In the formula, Indicates the commission ratio of live streaming merchants. Indicates the sales price of the product. Indicates the consumer's sensitivity to product quality cost, It represents the impact coefficient of live streaming on demand. represents the product quality cost, Indicates the consumer's sensitivity to the cost of live streaming marketing. The cost of live streaming marketing is Indicates the marketing cost ratio of live streaming merchants. It represents the proportion of product quality costs borne by live streaming merchants, and based on the profit function of live streaming merchants, the second-order derivative of live streaming marketing costs is obtained through the inverse solution method. The specific expression is as follows:

[0013] Therefore, it can be concluded that the profit function of live streaming merchants is a concave function of the live streaming marketing cost. is the profit function value of the live streaming salesperson, and the first-order derivative of the live streaming marketing cost is taken through the reverse solution method, and the result of the other-order derivative is zero, thus obtaining the optimal function 1 of the live streaming marketing cost, and its specific expression is as follows:

[0014] Then the optimal live broadcast marketing cost is determined. On the contrary, the live broadcast seller does not need to formulate a live broadcast sales strategy under the condition of sharing the product quality cost with the product manufacturer. The specific operation is that the product manufacturer first determines the sales price and product quality cost of the product, and the live broadcast seller decides the live broadcast marketing cost based on the sales price and product quality cost of the product. First, the profit function 2 of the live broadcast seller is constructed, and its specific expression is as follows:

[0015] In the formula, Indicates the sales price of the product. represents the product quality cost 2, The second optimal function of the live marketing cost is obtained by the same method as above. Its specific expression is as follows:

[0016] And then determine the optimal live streaming marketing cost.

[0017] Preferably, the product assisted sales unit includes a prediction and backtracking module and an assisted sales judgment module. The prediction and backtracking module is provided with a prediction and backtracking function. The specific operation is that the live broadcast sales merchant can modify the local topological features input into the product sales prediction model by modifying the product sales data during historical live broadcast sales, thereby making a prediction based on the modified local topological features through the product sales prediction model, and comparing the prediction result with the original prediction result to judge the impact of the modified local topological features on the original prediction result. The assisted sales judgment module and the prediction and backtracking module are electrically connected, and the assisted sales judgment module can modify the product sales and marketing strategy based on the impact of the modified local topological features on the original prediction results, thereby achieving assisted product sales.

[0018] The sales assistance method based on AI technology includes the following steps: S: Obtain the historical sales data of the products that the product manufacturer needs to sell through the live streaming merchants through the historical sales data collection module, process the acquired product historical sales data through the historical sales data processing module, convert the local topological features into vector form through the historical sales data vectorization module, and upload the local topological features converted into vector form to the system's built-in database for storage through the historical sales data storage module; S: Construct a product sales prediction model through the product sales prediction model construction module, construct a model training data set through the model training data set construction module, train the product sales prediction model through the product sales prediction model training module, and perform accuracy verification on the trained product sales prediction model, and obtain the product sales volume data output by the product sales prediction model at the prediction time selected by the live broadcast seller through the product sales prediction model output module; S: Construct a live streaming e-commerce supply chain through the live streaming e-commerce supply chain construction module, determine the live streaming marketing costs under different situations through the scenario sales strategy formulation module, judge the impact of modifying local topological features on the original prediction results through the prediction backtracking module, and modify the product sales and marketing strategy based on the impact of modifying local topological features on the original prediction results through the auxiliary sales judgment module.

[0019] Compared with the prior art, the beneficial effects of the present invention at least include: the present invention proposes a sales assistance system and method based on AI technology, the sales assistance system based on AI technology is applied to live streaming merchants to sell products for product manufacturers under a commission model, the product data collection unit is used to obtain product historical sales data, and the obtained product historical sales data is extracted and vectorized, so as to facilitate its use as a supplementary feature for subsequent model training, thereby improving the accuracy of the subsequent product sales prediction model, the product sales prediction model and the model training data set are constructed by the product sales prediction unit, and the product sales prediction model is trained and the accuracy verification operation is performed, and the product sales volume data at the prediction moment selected by the live streaming merchant is obtained through the trained product sales prediction model, the sales strategy formulation unit is used to decide the live streaming marketing cost of the live streaming merchant under two situations based on the prediction result of the product sales data, thereby improving the profit of the live streaming merchant from the aspect of reducing the live streaming marketing cost, and the prediction backtracking function in the product auxiliary sales unit is used to judge the impact of modifying the local topological features on the original prediction result, and modify the product sales marketing strategy to achieve auxiliary product sales. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention; Figure 2This is a schematic diagram of the process flow of a product data collection unit according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the product sales forecasting unit flow in an embodiment of the present invention; Figure 4 A schematic diagram of a sales strategy formulation unit flow chart for an embodiment of the present invention; Figure 5 The figure is a flow chart of a product auxiliary sales unit according to an embodiment of the present invention.

[0021] In the figure: 100, product data collection unit; 101, historical sales data collection module; 102, historical sales data processing module; 103, historical sales data quantization module; 104, historical sales data storage module; 200, product sales forecasting unit; 201, product sales forecasting model construction module; 202, model training data set construction module; 203, product sales forecasting model training module; 204, product sales forecasting model output module; 300, sales strategy formulation unit; 301, live streaming supply chain construction module; 302, situation-based sales strategy formulation module; 400, product auxiliary sales unit; 401, forecast backtracking module; 402, auxiliary sales judgment module. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] See also Figure 1-Figure 5 The present invention provides a technical solution: a sales assistance system based on AI technology, including a product data collection unit 100. The sales assistance system based on AI technology is applied to a live broadcasting merchant to sell products for a product manufacturer under a commission model. The product data collection unit 100 acquires the historical sales data of the products that the product manufacturer needs the live broadcasting merchant to sell, processes the acquired product historical sales data, and stores the processed product historical sales data; The product sales forecasting unit 200 is used for product sales volume forecasting. The product sales forecasting unit 200 constructs a product sales forecasting model and a model training data set. The product sales forecasting model is trained and its accuracy is verified through the model training data set. The product sales volume data at the forecasting moment selected by the live broadcasting merchant is obtained through the trained product sales forecasting model. The sales strategy formulation unit 300 builds a live streaming sales supply chain based on product manufacturers, live streaming sales providers, and consumers, and formulates live streaming sales strategies based on different situations based on product sales volume data at the predicted moment selected by the user; The product auxiliary sales unit 400 is provided with a prediction backtracking function, which makes modifications based on the original data input into the product sales prediction model, and compares the prediction result with the original prediction result to determine the impact on the original prediction result, thereby realizing product sales assistance.

[0024] The product data collection unit 100 includes a historical sales data collection module 101 and a historical sales data processing module 102. The historical sales data collection module 101 acquires the historical sales data of the products that the product manufacturer needs to sell through live streaming, including the sales data during the historical live streaming and the self-sales data of the network platform. The historical sales data processing module 102 is electrically connected to the historical sales data collection module 101. The historical sales data processing module 102 processes the acquired product historical sales data. The specific operation is to extract topological features from the product historical time series sales data without topological space structure through the Witness complex construction method, and in order to make the extracted topological features contain the data connection between the previous and next time, a local topological feature is calculated based on the product historical time series sales data on a sliding window of a size of n in a way of extracting local topological features, so that each local topological feature contains the information content of the evolution of the local topological feature in the time series historical sales data of size n over time, and the local topological feature is combined with the actual value in the corresponding product historical time series sales data through a persistent homology method based on the sliding window; The product data collection unit 100 also includes a historical sales data vectorization module 103 and a historical sales data storage module 104. The historical sales data vectorization module 103 is electrically connected to the historical sales data processing module 102. The historical sales data vectorization module 103 converts the local topological features into vector form by using a method of vectorizing the persistence graph by using kernel embedding, so that it can be subsequently input into the product sales prediction model for use. The historical sales data storage module 104 is electrically connected to the historical sales data vectorization module 103. The historical sales data storage module 104 uploads the local topological features converted into vector form to the system's built-in database for storage, and synchronously uploads the product historical time series sales data combined with the corresponding local topological features to the system's built-in database for storage; The product sales forecasting unit 200 includes a product sales forecasting model building module 201 and a model training data set building module 202. The product sales forecasting model building module 201 aims at product sales volume forecasting and builds a product sales forecasting model based on the Nbeats model. A multi-head attention mechanism is introduced on the basis of the original Nbeats model, so that the product sales forecasting model pays attention to the local topological features extracted from the product historical time series sales data. At the same time, the product sales forecasting model learns the topological structure contained in the product historical time series sales data, and uses the local topological features extracted from the product historical time series sales data as supplementary features for subsequent model training, thereby improving the prediction accuracy of the product sales forecasting model. The model training data set building module 202 crawls product sales data samples through a web crawler, extracts local topological features and performs vectorization processing on the crawled product sales data samples, and builds a model training data set based on the processed product sales data samples. The product sales forecasting unit 200 also includes a product sales forecasting model training module 203 and a product sales forecasting model output module 204. The product sales forecasting model training module 203 is electrically connected to the product sales forecasting model construction module 201 and the model training data set construction module 202 respectively. The product sales forecasting model training module 203 performs model training on the product sales forecasting model through the model training data set, and performs accuracy verification operation on the trained product sales forecasting model through the model accuracy algorithm. The specific calculation formula is as follows:

[0025] In the formula, represents the accuracy value of the product sales forecast model, Indicates the number of data samples predicted correctly, The number of data samples indicating prediction errors, the product sales prediction model output module 204 and the product sales prediction model training module 203 are electrically connected, the product sales prediction model output module 204 uses the local topological features in the form of vectors corresponding to the product sales data of the historical live streaming with goods stored in the database as the input of the product sales prediction model, and obtains the product sales volume data output by the product sales prediction model at the prediction moment selected by the live streaming with goods merchant; The sales strategy formulation unit 300 includes a live streaming goods supply chain construction module 301. The live streaming goods supply chain construction module 301 constructs a live streaming goods supply chain based on product manufacturers, live streaming goods merchants and consumers. Between product manufacturers and live streaming goods merchants, product manufacturers need to pay product sales commissions to live streaming goods merchants. Between product manufacturers and consumers, product manufacturers need to bear product quality costs. Between live streaming goods merchants and consumers, live streaming goods merchants need to bear live streaming marketing costs. The sales strategy formulation unit 300 also includes a situation-based sales strategy formulation module 302, which is electrically connected to the live streaming supply chain construction module 301. The situation-based sales strategy formulation module 302 outputs the product sales volume data at the prediction time selected by the live streaming merchant based on the product sales prediction model, and compares it with the product sales threshold set by the live streaming merchant. When the product sales volume data output by the product sales prediction model at the prediction time selected by the live streaming merchant is greater than the product sales threshold set by the live streaming merchant, the live streaming merchant needs to formulate a live streaming sales strategy while sharing the product quality cost with the product manufacturer. The specific operation is that the live streaming merchant first determines the proportion of the product quality cost to be borne, and the product manufacturer then determines the sales price and product quality cost of the product. The live streaming merchant finally decides the live streaming marketing cost based on the sales price and product quality cost of the product. First, a profit function 1 of the live streaming merchant is constructed, and its specific expression is as follows:

[0026] In the formula, Indicates the commission ratio of live streaming merchants. Indicates the sales price of the product. Indicates the consumer's sensitivity to product quality cost, It represents the impact coefficient of live streaming on demand. represents the product quality cost, Indicates the consumer's sensitivity to the cost of live streaming marketing. The cost of live streaming marketing is Indicates the marketing cost ratio of live streaming merchants. Indicates the proportion of product quality costs borne by live streaming merchants. It is expressed as the marketing cost ratio of the live broadcast room, and based on the profit function of the live broadcast merchant, the second-order derivative of the live broadcast marketing cost is obtained through the inverse solution method. The specific expression is as follows:

[0027] Therefore, it can be concluded that the profit function of live streaming merchants is a concave function of the live streaming marketing cost. is the profit function value of the live streaming salesperson, and the first-order derivative of the live streaming marketing cost is taken through the reverse solution method, and the result of the other-order derivative is zero, thus obtaining the optimal function 1 of the live streaming marketing cost, and its specific expression is as follows:

[0028] Then the optimal live broadcast marketing cost is determined. On the contrary, the live broadcast seller does not need to formulate a live broadcast sales strategy under the condition of sharing the product quality cost with the product manufacturer. The specific operation is that the product manufacturer first determines the sales price and product quality cost of the product, and the live broadcast seller decides the live broadcast marketing cost based on the sales price and product quality cost of the product. First, the profit function 2 of the live broadcast seller is constructed, and its specific expression is as follows:

[0029] In the formula, Indicates the sales price of the product. represents the product quality cost 2, The second optimal function of the live marketing cost is obtained by the same method as above. Its specific expression is as follows:

[0030] Then determine the optimal live broadcast marketing cost; The product auxiliary sales unit 400 includes a prediction and backtracking module 401 and an auxiliary sales judgment module 402. The prediction and backtracking module 401 is provided with a prediction and backtracking function. The specific operation is that the live broadcast seller can modify the local topological features input into the product sales prediction model by modifying the product sales data during the historical live broadcast, so as to make a prediction based on the modified local topological features through the product sales prediction model, and compare the prediction result with the original prediction result, so as to judge the influence of the modified local topological features on the original prediction result. The auxiliary sales judgment module 402 is electrically connected to the prediction and backtracking module 401. The auxiliary sales judgment module 402 can modify the product sales and marketing strategy based on the influence of the modified local topological features on the original prediction result, so as to achieve auxiliary product sales; The sales assistance method based on AI technology includes the following steps: S1: The historical sales data of the products that the product manufacturer needs to sell through the live streaming merchants are obtained through the historical sales data collection module 101, the historical sales data processing module 102 processes the obtained historical sales data of the products, the local topological features are converted into vector form through the historical sales data vectorization module 103, and the local topological features converted into vector form are uploaded to the system's built-in database for storage through the historical sales data storage module 104; S2: construct a product sales prediction model through the product sales prediction model construction module 201, construct a model training data set through the model training data set construction module 202, perform model training on the product sales prediction model through the product sales prediction model training module 203, and perform accuracy verification operation on the trained product sales prediction model, and obtain the product sales volume data output by the product sales prediction model at the prediction time selected by the live broadcast seller through the product sales prediction model output module 204; S3: Construct a live streaming e-commerce supply chain through the live streaming e-commerce supply chain construction module 301, determine the live streaming marketing costs under different situations through the situation-based sales strategy formulation module 302, judge the impact of modifying local topological features on the original prediction results through the prediction backtracking module 401, and modify the product sales and marketing strategy based on the impact of modifying local topological features on the original prediction results through the auxiliary sales judgment module 402.

[0031] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0032] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The sales assistance system based on AI technology is characterized by: The invention comprises a product data collection unit (100), wherein the sales assistance system based on AI technology is applied to a live broadcast salesperson to sell products for a product manufacturer under a commission model, wherein the product data collection unit (100) acquires historical sales data of products that the product manufacturer needs the live broadcast salesperson to sell, processes the acquired historical sales data of the products, and stores the processed historical sales data of the products; A product sales prediction unit (200), wherein the product sales prediction unit (200) is for the purpose of predicting product sales volume, constructs a product sales prediction model and a model training data set, trains the product sales prediction model and verifies its accuracy through the model training data set, and obtains product sales volume data at a prediction moment selected by a live broadcast salesperson through the trained product sales prediction model; A sales strategy formulation unit (300), wherein the sales strategy formulation unit (300) constructs a live streaming sales supply chain based on product manufacturers, live streaming sales providers and consumers, and formulates live streaming sales strategies based on different situations based on product sales volume data at a predicted time selected by a user; A product sales assistance unit (400) is provided with a prediction backtracking function, which modifies the original data input into the product sales prediction model, compares the prediction result with the original prediction result, thereby determining the impact on the original prediction result, and further realizing product sales assistance.

2. The AI-based sales support system according to claim 1, characterized in that: The product data collection unit (100) includes a historical sales data collection module (101) and a historical sales data processing module (102). The historical sales data collection module (101) acquires historical sales data of products that the product manufacturer needs to be sold by live streaming merchants, including historical sales data during live streaming and self-sales data on the network platform. The historical sales data processing module (102) is electrically connected to the historical sales data collection module (101). The historical sales data processing module (102) processes the acquired product historical sales data. The specific operation is to use a Witness complex to construct The method extracts topological features from product historical time series sales data that have no topological spatial structure. In order to make the extracted topological features contain the data connection between the previous and next time, a local topological feature is calculated based on the product historical time series sales data on a sliding window with a size of n by extracting local topological features, so that each local topological feature contains the information content of the local topological feature in the time series historical sales data with a size of n that evolves over time, and the local topological features are combined with the actual values ​​in the corresponding product historical time series sales data through a persistent homology method based on a sliding window.

3. The AI-based sales support system according to claim 2, characterized in that: The product data collection unit (100) further comprises a historical sales data vectorization module (103) and a historical sales data storage module (104), wherein the historical sales data vectorization module (103) and the historical sales data processing module (102) are electrically connected, wherein the historical sales data vectorization module (103) converts local topological features into vector form by using a method of vectorizing a persistence graph using kernel embedding, so that the local topological features can be subsequently input into a product sales prediction model for use, wherein the historical sales data storage module (104) and the historical sales data vectorization module (103) are electrically connected, wherein the historical sales data storage module (104) uploads the local topological features converted into vector form into a system built-in database for storage, and simultaneously uploads the product historical time series sales data combined with the corresponding local topological features into the system built-in database for storage.

4. The AI-based sales support system according to claim 1, characterized in that: The product sales prediction unit (200) comprises a product sales prediction model construction module (201) and a model training data set construction module (202). The product sales prediction model construction module (201) is for the purpose of product sales volume prediction and constructs a product sales prediction model based on the Nbeats model. A multi-head attention mechanism is introduced on the basis of the original Nbeats model, so that the product sales prediction model pays attention to the local topological features extracted from the product historical time series sales data. At the same time, the product sales prediction model learns the topological structure contained in the product historical time series sales data. The local topological features extracted from the product historical time series sales data are used as supplementary features for subsequent model training, thereby improving the prediction accuracy of the product sales prediction model. The model training data set construction module (202) crawls product sales data samples through a web crawler, extracts local topological features and performs vectorization processing on the crawled product sales data samples, and constructs a model training data set based on the processed product sales data samples.

5. The AI-based sales support system according to claim 4, characterized in that: The product sales forecasting unit (200) further includes a product sales forecasting model training module (203) and a product sales forecasting model output module (204). The product sales forecasting model training module (203) is electrically connected to the product sales forecasting model construction module (201) and the model training data set construction module (202), respectively. The product sales forecasting model training module (203) performs model training on the product sales forecasting model through the model training data set, and performs accuracy verification operation on the trained product sales forecasting model through the model accuracy algorithm. The product sales forecasting model output module (204) is electrically connected to the product sales forecasting model training module (203). The product sales forecasting model output module (204) uses the local topological features in the form of vectors corresponding to the product sales data of the historical live broadcast with goods stored in the database as the input of the product sales forecasting model, and obtains the product sales volume data output by the product sales forecasting model at the prediction time selected by the live broadcast with goods merchant.

6. The AI-based sales support system according to claim 1, characterized in that: The sales strategy formulation unit (300) includes a live streaming e-commerce supply chain construction module (301), which constructs a live streaming e-commerce supply chain based on product manufacturers, live streaming e-commerce merchants and consumers. Between product manufacturers and live streaming e-commerce merchants, product manufacturers need to pay product sales commissions to live streaming e-commerce merchants, between product manufacturers and consumers, product manufacturers need to bear product quality costs, and between live streaming e-commerce merchants and consumers, live streaming e-commerce merchants need to bear live streaming marketing costs.

7. The AI-based sales support system according to claim 6, characterized in that: The sales strategy formulation unit (300) also includes a situation-based sales strategy formulation module (302), and the situation-based sales strategy formulation module (302) is electrically connected to the live streaming e-commerce supply chain construction module (301). The situation-based sales strategy formulation module (302) is based on the product sales volume data output by the product sales forecasting model at the forecast time selected by the live streaming e-commerce merchant, and compares it with the product sales threshold set by the live streaming e-commerce merchant. When the product sales volume data output by the product sales forecasting model at the forecast time selected by the live streaming e-commerce merchant is greater than the product sales threshold set by the live streaming e-commerce merchant, the live streaming e-commerce merchant needs to formulate a live streaming sales strategy while sharing the product quality cost with the product manufacturer. Otherwise, the live streaming e-commerce merchant does not need to formulate a live streaming sales strategy while sharing the product quality cost with the product manufacturer.

8. The AI-based sales support system according to claim 1, characterized in that: The product auxiliary sales unit (400) includes a prediction and backtracking module (401) and an auxiliary sales judgment module (402). The prediction and backtracking module (401) is provided with a prediction and backtracking function. The specific operation is that the live broadcast sales merchant can modify the local topological features input into the product sales prediction model by modifying the product sales data during the historical live broadcast sales, so as to make a prediction based on the modified local topological features through the product sales prediction model, and compare the prediction result with the original prediction result to judge the impact of the modified local topological features on the original prediction result. The auxiliary sales judgment module (402) and the prediction and backtracking module (401) are electrically connected. The auxiliary sales judgment module (402) can modify the product sales and marketing strategy based on the impact of the modified local topological features on the original prediction result, thereby realizing auxiliary product sales.

9. The sales assistance method based on AI technology is characterized by The following steps are involved: S1: Obtaining historical sales data of products that a product manufacturer needs to sell through a live streaming salesperson through a historical sales data collection module (101), processing the acquired historical sales data of the products through a historical sales data processing module (102), converting local topological features into vector form through a historical sales data vectorization module (103), and uploading the local topological features converted into vector form into a system built-in database for storage through a historical sales data storage module (104); S2: constructing a product sales prediction model through a product sales prediction model construction module (201), constructing a model training data set through a model training data set construction module (202), training the product sales prediction model through a product sales prediction model training module (203), and performing an accuracy verification operation on the trained product sales prediction model, and obtaining product sales volume data output by the product sales prediction model at the prediction time selected by the live broadcast sales provider through a product sales prediction model output module (204); S3: Construct a live streaming e-commerce supply chain through the live streaming e-commerce supply chain construction module (301), determine the live streaming marketing costs under different situations through the situation-based sales strategy formulation module (302), judge the impact of modifying local topological features on the original prediction results through the prediction backtracking module (401), and modify the product sales and marketing strategy based on the impact of modifying local topological features on the original prediction results through the auxiliary sales judgment module (402).

Images