Commodity recommendation method and system based on artificial intelligence

Abstract

The invention discloses a commodity recommendation method and system based on artificial intelligence, and the method comprises the steps: judging the seasonal sensitivity types of a user in different seasons according to the purchase behavior data of the user in each season; determining a seasonal commodity purchasing habit of the user according to the user behavior data of the user in different seasons; determining the commodity preference of the user through the user behavior data, and generating a preliminary commodity recommendation list of each season; identifying high-frequency commodity combinations purchased by the user in different seasons, and generating a commodity recommendation list of subsequent seasons; calculating a correlation score and a diversity score of each commodity in the commodity recommendation list in the current season, determining an optimal recommended commodity set, and adjusting the commodity recommendation list; and monitoring a commodity recommendation effect in real time, and adjusting the commodity recommendation list according to the commodity recommendation effect. According to the invention, an efficient, accurate and dynamically adjusted commodity recommendation strategy is provided, and the sales conversion rate of the e-commerce platform and the shopping experience of the user are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a commodity recommendation method and system based on artificial intelligence. Background Art

[0002] With the rapid development of e-commerce platforms, product recommendation systems have become an important tool for improving user shopping experience and platform sales conversion rate. However, existing recommendation systems have many shortcomings in dealing with users' complex purchasing behaviors and seasonal needs, and are in urgent need of further optimization. Traditional recommendation systems are mainly based on users' historical purchase records and click behaviors, and lack in-depth analysis of seasonal factors. This method makes it difficult to accurately identify seasonal products, resulting in the inability to effectively recommend products that users currently need in a specific season. For example, recommending winter clothing in summer or recommending summer sunscreen products in winter will significantly reduce user satisfaction. Different users have significant differences in sensitivity to seasonal products. Some users may show a high demand for specific products in certain seasons, while other users may have less demand for these products. Existing systems are usually unable to distinguish such differences among users, resulting in the recommended products not really meeting users' seasonal needs, which in turn affects user experience and platform sales conversion rate. Users' purchasing behavior is not only affected by seasons, but also closely related to personal habits, lifestyles and other factors. Traditional recommendation algorithms often lack in-depth analysis of users' purchasing habits, resulting in the recommended product combinations being not personalized and accurate enough, making it difficult to meet users' diverse purchasing needs. For example, for users who are used to buying seasonal goods before the season begins, traditional recommendation systems cannot recommend suitable goods in advance. Current recommendation systems usually focus on recommending goods that are similar to the user's historical behavior, resulting in insufficient diversity in the recommended content. This single recommendation model can easily cause visual fatigue for users, reducing their interest and stickiness to the platform. At the same time, the lack of diverse recommendations will also miss the opportunity to show users more potential products of interest, limiting the platform's sales growth. Users' purchasing behavior is affected by external factors, such as seasonal promotions, changes in personal life, etc., which may cause user behavior patterns to shift. Traditional recommendation systems lack the ability to monitor and adjust in real time, making it difficult to capture changes in user behavior in a timely manner, which in turn affects the accuracy and timeliness of recommendations. When processing product attribute data, existing recommendation systems usually only consider a few significant attributes and fail to fully analyze the similarities and differences between products. This deficiency affects the performance of the recommendation system in balancing product relevance and diversity, and is unable to provide users with a rich and diverse selection of products. The existence of these problems has led to obvious deficiencies in the current recommendation system in terms of user experience and recommendation effect. Users often cannot get product recommendations that meet their current needs, especially during seasonal changes or special shopping festivals. In order to improve the accuracy of recommendations and user satisfaction, there is an urgent need for an advanced recommendation method that can effectively identify seasonal products, capture user seasonal sensitivity, mine purchasing habits and preferences, and dynamically adjust recommendation strategies. Summary of the invention

[0003] The present invention solves the problems existing in the above-mentioned prior art and provides a commodity recommendation method based on artificial intelligence, which mainly includes:

[0004] Determine whether a product is a seasonal product based on its sales time, sales volume, and sales fluctuation index. Calculate the user's seasonal sensitivity based on the user's purchase behavior data in each season and determine the user's seasonal sensitivity type in different seasons.

[0005] Determine the user's seasonal product purchasing habits based on the user's behavior data in different seasons;

[0006] Determine the user's product preferences through user behavior data, generate a preliminary product recommendation list for each season, adjust the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determine the timing of product recommendations in different seasons based on the user's purchase habit classification results;

[0007] Based on the user's purchase record data for the current season and the historical purchase record data for the season, determine whether the user has a purchase behavior deviation, identify the high-frequency product combinations purchased by the user in different seasons, and generate a product recommendation list for subsequent seasons;

[0008] Based on the sales data of goods in different seasons, user purchase behavior data and attribute data of goods, the relevance score and diversity score of each product in the product recommendation list in the current season are calculated. Based on the multi-objective optimization function, the seasonal relevance and product diversity of the recommended goods in the product recommendation list are balanced, the best recommended product set is determined, and the product recommendation list is adjusted.

[0009] Monitor the product recommendation effect in real time and adjust the product recommendation list based on the product recommendation effect.

[0010] Further, judging whether a product is a seasonal product based on the sales time, sales volume and sales fluctuation index of the product, and calculating the seasonal sensitivity of the user based on the purchase behavior data of the user in each season, and judging the seasonal sensitivity type of the user in different seasons, include:

[0011] The sales time and sales volume of the goods are extracted from the e-commerce platform database, and the obtained data is cleaned to remove outliers and missing values, and the date field is converted into a time series format; the moving average method is used to calculate sales fluctuations to obtain sales fluctuation indicators of the goods, including sales peak and valley values, standard deviation and variance of sales, and seasonal index. The seasonal index is the ratio of the sales volume of the goods in different seasons to the average sales volume; based on the sales volume and sales fluctuation indicators of the goods, the K-means clustering algorithm is used for model training to determine whether the goods are seasonal goods; the purchase record data of seasonal goods purchased by users in different seasons are extracted from the sales records in the e-commerce platform database. The purchase record data includes product ID, product category, purchase time, purchase quantity and purchase amount; the user's seasonal product purchase record data is divided by season to determine the user's purchase behavior data in each season. The purchase behavior data includes purchase frequency, purchase volume and purchase amount; based on the user's purchase behavior data in each season, the seasonal sensitivity evaluation formula is used Calculate the seasonal sensitivity of the user, where F represents the purchase frequency of seasonal products, Q represents the purchase volume of seasonal products, A represents the purchase amount of seasonal products, and F represents the purchase frequency of seasonal products. Total represents the total purchase frequency of all commodities in the season, Q Total A represents the total purchase volume of all commodities in this season. Total represents the total purchase amount of all commodities in the season, S represents the seasonal sensitivity of the user, χ represents the influence of purchase frequency, δ represents the influence of purchase volume, ε represents the influence of purchase amount, and φ represents the index adjustment factor, which is used to control the overall influence of purchase frequency, purchase volume and purchase amount when calculating seasonal sensitivity. χ, δ, ε and φ are all obtained by exploratory analysis of historical purchase data. According to the calculated seasonal sensitivity, the sensitivity threshold is set to determine the seasonal sensitivity type of the user in different seasons, including high seasonal sensitivity and low seasonal sensitivity.

[0012] Furthermore, the method of determining the user's seasonal commodity purchasing habits based on the user's behavior data in different seasons includes:

[0013] Through the e-commerce platform database, the user behavior data of users in different seasons are queried based on the user ID, and the user's seasonal sensitivity type is marked. The user behavior data includes browsing records, click records and purchase records; according to the preset time period, the user behavior data of users in different seasons are divided into different time intervals, and the user behavior data are further divided according to the classification labels of the goods. The classification labels of the goods include seasonal goods and non-seasonal goods; the user behavior data are grouped according to user characteristics, and user characteristics include age, gender, and geographical location; according to the purchase frequency, purchase quantity and purchase amount of different commodity categories of each user in each time period of different seasons, the decision tree algorithm is used for model training to determine the user's seasonal commodity purchasing habits, which include advance purchase, immediate purchase and delayed purchase.

[0014] Furthermore, the method of determining the user's product preferences through the user behavior data, generating a preliminary product recommendation list for each season, adjusting the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determining the product recommendation timing in different seasons in combination with the user's purchase habit classification results, includes:

[0015] Determine the user's product preferences based on user behavior data; use a product-based collaborative filtering recommendation algorithm to calculate the similarity between products based on the user's product preferences in each season, and use the similarity matrix to generate a preliminary product recommendation list for each season; use the user's purchase behavior data in each season and use the seasonal sensitivity evaluation formula to obtain the user's seasonal sensitivity in each season; if the user is a high seasonal sensitivity user, use the seasonal product recommendation ratio adjustment formula P1=P×(1+S) based on the user's seasonal sensitivity to determine the final recommendation ratio of seasonal products, adjust the seasonal product recommendation ratio in the preliminary recommendation list, and generate a product recommendation list for each season, where P1 is the adjusted final recommendation ratio of seasonal products, and P is the preset basic recommendation ratio of seasonal products; adjust the recommendation timing of the product recommendation list based on the user's seasonal product purchasing habits.

[0016] Furthermore, judging whether the user has a purchase behavior deviation based on the user's purchase record data of the current season and the historical purchase record data of the season, identifying the high-frequency product combinations purchased by the user in different seasons, and generating a product recommendation list for subsequent seasons includes:

[0017] Through the e-commerce platform database, the user's purchase record data of the current season and the historical purchase record data of the season are obtained; the similarity between the user's purchase record data of the current season and the historical purchase record data of the season is calculated through the cosine similarity method, and it is determined whether the similarity is lower than the preset similarity threshold; if the similarity is lower than the preset similarity threshold, it is determined that the user has a purchase behavior deviation; if the user has a purchase behavior deviation, the actual clicks and purchased products of the user in the current season are obtained, the product recommendation list is adjusted, and the user's purchase record data in different seasons are obtained and organized into a transaction data set format, where each transaction represents a purchase record; a support threshold is set, and the sample is collected. The Apriori algorithm is used to mine frequent itemsets, and the high-frequency product combinations purchased by users in different seasons are identified based on the support threshold. The high-frequency product combinations are product combinations that appear at a frequency higher than the preset frequency threshold. Based on the high-frequency product combinations and their support, the FP-Growth algorithm is used to identify the association rules between the high-frequency product combinations, calculate the confidence and lift, and establish a product association database. The association rules and their support, confidence and lift are stored in the product association database, and the association rules of cross-season products are recorded. Based on the high-frequency product combinations purchased in different seasons and the association rules of cross-season products, a recommendation list of products for subsequent seasons is generated.

[0018] Further, the relevance score and diversity score of each product in the product recommendation list in the current season are calculated based on the product sales data, user purchase behavior data and product attribute data of different seasons, and the seasonal relevance and product diversity of the recommended products in the product recommendation list are balanced according to the multi-objective optimization function, the best recommended product set is determined, and the product recommendation list is adjusted, including:

[0019] Through the e-commerce platform database, obtain the sales data of goods in different seasons, user purchase behavior data and attribute data of goods, including price, brand and function; according to the seasonal correlation indicator formula Calculate the relevance score of each product in the product recommendation list in the current season, where R ij is the sales volume of product i in season j, W j is the weight of season j, N is the total number of seasons; according to the product diversity index formula Calculate the diversity score of the recommended product set in the current product recommendation list, where M is the total number of products and S is the kl is the similarity between product k and product l, which is obtained by using the cosine similarity calculation method based on the attribute data of the products; according to the multi-objective optimization function Balance the seasonal relevance and product diversity of the recommended products in the product recommendation list, and determine the best recommended product set, where α and β are weight parameters used to adjust the balance between relevance and diversity; adjust the product recommendation list according to the best recommended product collection, and dynamically update the seasonal relevance and product diversity data, and adjust the product recommendation list based on the real-time calculation results of the multi-objective optimization function F.

[0020] Furthermore, the real-time monitoring of the product recommendation effect and adjusting the product recommendation list according to the product recommendation effect include:

[0021] Obtain real-time product recommendation monitoring data through the e-commerce platform database, compare the recommended products with the products actually clicked and purchased by users, and determine the recommendation accuracy; record the user's behavior data of clicking on the recommended products and calculate the user's click rate; calculate the purchase conversion rate by comparing the user's behavior data of clicking on the recommended products with the actual purchase of the products; judge whether the product recommendation effect meets the expectations based on the analysis results of the recommendation accuracy, user click rate and purchase conversion rate, as well as the preset product recommendation evaluation standards, including the target values ​​of the recommendation accuracy, user click rate and purchase conversion rate; if the product recommendation effect does not meet the expectations, adjust the product recommendation list based on the users' actual clicks and purchases.

[0022] The second aspect of the present invention provides a product recommendation system based on artificial intelligence, which mainly includes:

[0023] Seasonal sensitivity analysis module, which is used to determine whether a product is a seasonal product based on the product's sales time, sales volume, and sales fluctuation index, and calculate the user's seasonal sensitivity based on the user's purchase behavior data in each season, and determine the user's seasonal sensitivity type in different seasons;

[0024] User purchasing habit analysis module, used to determine the user's seasonal commodity purchasing habits based on the user's behavior data in different seasons;

[0025] The product recommendation module is used to determine the user's product preferences through user behavior data, generate a preliminary product recommendation list for each season, adjust the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determine the timing of product recommendations in different seasons based on the user's purchase habit classification results;

[0026] The cross-seasonal product recommendation module is used to determine whether the user has a purchase behavior deviation based on the user's current season purchase record data and the historical purchase record data of the season, and identify the high-frequency product combinations purchased by the user in different seasons, and generate a product recommendation list for the subsequent season;

[0027] The product recommendation adjustment module is used to calculate the relevance score and diversity score of each product in the product recommendation list in the current season based on the product sales data, user purchase behavior data and product attribute data of different seasons, and balance the seasonal relevance and product diversity of the recommended products in the product recommendation list according to the multi-objective optimization function, determine the best recommended product set, and adjust the product recommendation list;

[0028] The product recommendation effect monitoring module is used to monitor the product recommendation effect in real time and adjust the product recommendation list according to the product recommendation effect.

[0029] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0030] The present invention provides a commodity recommendation method and system based on artificial intelligence. By analyzing the sales time, sales volume and sales fluctuation index of the commodity, the present invention can accurately determine whether the commodity is a seasonal commodity, thereby ensuring that the recommended commodity is highly matched with the actual needs of users in different seasons, and improving the relevance and effectiveness of the recommendation. Based on the purchase behavior data of users in each season, the seasonal sensitivity of the user is calculated, and the seasonal sensitivity type is determined, so that the consumption behavior and needs of the user in different seasons can be personalized and identified, and user satisfaction can be significantly improved. The present invention determines the user's commodity preferences and the purchase habits of seasonal commodities by analyzing the user behavior data, generates a preliminary commodity recommendation list for each season, and adjusts the recommendation ratio and recommendation timing of seasonal commodities based on the user's seasonal sensitivity, so that the recommendation is more personalized and accurate. For the real-time monitoring and dynamic adjustment of user purchase behavior, the present invention can effectively identify user behavior deviation, timely adjust the recommendation strategy, and ensure that the recommended content can adapt to the changes in user needs. At the same time, by calculating the relevance score and diversity score of the recommended commodity, and using the multi-objective optimization function to balance the seasonal relevance and diversity of the recommended commodity, the present invention can provide the best recommended commodity set, avoid the simplification of the recommended content, and enhance user interest and platform stickiness. Monitor the product recommendation effect in real time and adjust the recommendation list according to the results to ensure that the recommendation system can be continuously optimized to improve the recommendation accuracy and user satisfaction. This invention provides an efficient, accurate and dynamically adjusted product recommendation strategy by comprehensively considering the seasonality of products, user behavior habits and diverse needs, which significantly improves the sales conversion rate and user shopping experience of the e-commerce platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flowchart of a commodity recommendation method based on artificial intelligence of the present invention;

[0032] Figure 2 A schematic diagram of a commodity recommendation system based on artificial intelligence of the present invention;

[0033] Figure 3 A schematic diagram of a commodity recommendation method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 and Figure 3 In this embodiment, a commodity recommendation method based on artificial intelligence may specifically include:

[0036] Step S101, judging whether the commodity is a seasonal commodity according to the sales time, sales volume and sales fluctuation index of the commodity, and calculating the seasonal sensitivity of the user according to the purchasing behavior data of the user in each season, and judging the seasonal sensitivity type of the user in different seasons.

[0037] The sales time and sales volume of the goods are extracted from the e-commerce platform database, and the acquired data is cleaned to remove outliers and missing values, and the date field is converted to a time series format. The moving average method is used to calculate sales fluctuations to obtain sales fluctuation indicators of the goods, including sales peaks and valleys, standard deviations and variances of sales, and seasonality indexes. The seasonality index is the ratio of the sales volume of the goods in different seasons to the average sales volume. According to the sales volume and sales fluctuation indicators of the goods, the K-means clustering algorithm is used for model training to determine whether the goods are seasonal goods. The purchase record data of seasonal goods purchased by users in different seasons are extracted from the sales records of the e-commerce platform database. The purchase record data includes product ID, product category, purchase time, purchase quantity and purchase amount. The user's seasonal product purchase record data is divided by season to determine the user's purchase behavior data in each season. The purchase behavior data includes purchase frequency, purchase volume and purchase amount. According to the user's purchase behavior data in each season, the seasonal sensitivity evaluation formula is used Calculate the seasonal sensitivity of the user, where F represents the purchase frequency of seasonal products, Q represents the purchase volume of seasonal products, A represents the purchase amount of seasonal products, and F represents the purchase frequency of seasonal products. Total represents the total purchase frequency of all commodities in the season, Q Total A represents the total purchase volume of all commodities in this season. Totalrepresents the total purchase amount of all commodities in the season, S represents the seasonal sensitivity of the user, χ represents the influence of purchase frequency, δ represents the influence of purchase volume, ε represents the influence of purchase amount, and φ represents the index adjustment factor, which is used to control the overall influence of purchase frequency, purchase volume, and purchase amount when calculating seasonal sensitivity. χ, δ, ε, and φ are all obtained through exploratory analysis of historical purchase data. Based on the calculated seasonal sensitivity, the sensitivity threshold is set to determine the seasonal sensitivity type of the user in different seasons, including high seasonal sensitivity and low seasonal sensitivity.

[0038] Exemplarily, the sales time and sales volume data of a certain product are extracted from the e-commerce platform database. Through data cleaning, outliers and missing values ​​are removed, and the date field is converted into a time series format, and the data is obtained as (2023-01-01, 120), (2023-01-02, 85), (2023-01-03, 95), ... (2023-12-31, 150). The 7-day moving average method is used to calculate sales fluctuations, and the moving average value of each day is obtained. For January 8, 2023, its moving average value is the sales volume average value of the previous 7 days, from January 1, 2023 to January 7, 2023. The peak and valley values ​​of sales volume are identified by the moving average value, and the peak and valley values ​​of sales volume are 200 pieces and 50 pieces respectively on June 15, 2023 and December 20, 2023. The standard deviation and variance of the sales volume of the entire time series data are calculated, and the standard deviation is 30 and the variance is 900. The ratio of the sales volume of goods in different seasons to the average sales volume is calculated, and the seasonal indexes of spring, summer, autumn and winter are 1.2, 0.8, 1.1 and 0.9 respectively, which means that the sales volume in spring is 20% higher than the average sales volume, 20% lower in summer, 10% higher in autumn and 10% lower in winter. The sales volume and sales fluctuation index of each product are used as feature inputs, and the K-means clustering algorithm is used to cluster the products. Two cluster centers are selected. After model training, the products are divided into two categories, one of which is seasonal products and the other is non-seasonal products. The seasonal product purchase records of a user are extracted from the database of the e-commerce platform. In spring, the user purchased gardening tools with a purchase frequency of 5 times, a total purchase of 10 pieces, and a total amount of 500 yuan. In summer, the user purchased sunscreen with a purchase frequency of 7 times, a total purchase of 14 bottles, and a total amount of 280 yuan. In autumn, the user purchased sweaters with a purchase frequency of 4 times, a total purchase of 8 pieces, and a total amount of 1,200 yuan. In winter, the user purchased a heater with a purchase frequency of 3 times, a total purchase volume of 3 units, and a total amount of 1,500 yuan. The total purchase frequency, total purchase volume, and total purchase amount of all products of the user in this season are obtained, and the total purchase frequency in spring is 100 times, the total purchase volume is 200 pieces, and the total amount is 10,000 yuan. The total purchase frequency in summer is 150 times, the total purchase volume is 300 bottles, and the total amount is 12,000 yuan. The total purchase frequency in autumn is 80 times, the total purchase volume is 160 pieces, and the total amount is 16,000 yuan. The total purchase frequency in winter is 50 times, the total purchase volume is 50 units, and the total amount is 25,000 yuan. According to the seasonal sensitivity evaluation formula Calculate the seasonal sensitivity of the user, where F represents the purchase frequency of seasonal products, Q represents the purchase volume of seasonal products, A represents the purchase amount of seasonal products, and F represents the purchase frequency of seasonal products. Total represents the total purchase frequency of all commodities in the season, QTotal A represents the total purchase volume of all commodities in this season. Total represents the total purchase amount of all commodities in the season, S represents the seasonal sensitivity of the user, χ represents the influence of purchase frequency, δ represents the influence of purchase volume, ε represents the influence of purchase amount, and φ represents the index adjustment factor, which is used to control the overall influence of purchase frequency, purchase volume, and purchase amount in calculating seasonal sensitivity. χ, δ, ε, and φ are all obtained by exploratory analysis of historical purchase data. The introduction of the exponential growth part aims to capture the extremely high or low purchase frequency, purchase volume, and purchase amount in a certain season, which can effectively reflect that some users have a strong preference for a certain type of commodity in a specific season. If this preference is not properly amplified nonlinearly, it may be diluted in a purely linear weighted process. If γ = 0.5, δ = 0.3, and ε = 0.2, the user's spring seasonal sensitivity S is calculated to be 0.0581, the summer seasonal sensitivity S is 0.0457, the autumn seasonal sensitivity S is 0.0646, and the winter seasonal sensitivity S is 0.0718. By setting a sensitivity threshold, such as 0.06, the seasonal sensitivity types of the user in different seasons are determined, including high seasonal sensitivity and low seasonal sensitivity. It can be determined that the user has high seasonal sensitivity in autumn and winter.

[0039] Step S102, determining the user's seasonal commodity purchasing habits based on the user's behavior data in different seasons.

[0040] Through the e-commerce platform database, the user behavior data of users in different seasons are queried based on the user ID, and the seasonal sensitivity type of the user is marked. The user behavior data includes browsing records, click records and purchase records. According to the preset time period, the user behavior data of users in different seasons are divided into different time intervals, and the user behavior data is further divided according to the classification labels of the goods. The classification labels of the goods include seasonal goods and non-seasonal goods. The user behavior data is grouped according to user characteristics, which include age, gender, and geographic location. According to the purchase frequency, purchase quantity and purchase amount of different commodity categories of each user in different time periods in different seasons, the decision tree algorithm is used for model training to determine the user's seasonal commodity purchasing habits, which include advance purchase, immediate purchase and delayed purchase.

[0041] For example, in the database of an e-commerce platform, the behavior data of a user in different seasons is queried through the user ID. The user often browses and clicks on warm-keeping products, including down jackets and heaters, during the winter, and makes multiple purchases in early January, that is, in winter. In data processing, the user behavior data is divided into different time intervals, December to February of the following year is winter, and these products are marked as seasonal products. The user also browsed some non-seasonal products such as books and electronic products during this period, and these data were also classified and sorted. According to the user's age of 30 years old, gender of female, and geographical location of northern cities, the user's behavior data is grouped and analyzed with other user data with similar characteristics. By analyzing the user's purchase records in winter, it is identified that the user has begun to browse and click on a large number of warm-keeping products in early December, and has made multiple purchases in early January. This shows that the user tends to buy seasonal products needed for winter in advance, such as down jackets and heaters. After model training using the decision tree algorithm, it is determined that the user's purchasing habit type is advance purchase, and the purchasing habits include advance purchase, immediate purchase, and delayed purchase. This behavior suggests that the user is actively purchasing seasonal items before or at the beginning of the season, rather than waiting for the middle or end of the season.

[0042] Step S103, determine the user's product preference through user behavior data, generate a preliminary product recommendation list for each season, adjust the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determine the timing of product recommendation in different seasons based on the user's purchase habit classification results.

[0043] Determine the user's product preferences based on user behavior data. Based on the user's product preferences in each season, use a product-based collaborative filtering recommendation algorithm to calculate the similarity between products, and use the similarity matrix to generate a preliminary product recommendation list for each season. Through the user's purchase behavior data in each season, use the seasonal sensitivity evaluation formula to obtain the user's seasonal sensitivity in each season. If the user is a user with high seasonal sensitivity, use the seasonal product recommendation ratio adjustment formula P1=P×(1+S) based on the user's seasonal sensitivity to determine the final recommendation ratio of seasonal products, adjust the seasonal product recommendation ratio in the preliminary recommendation list, and generate a product recommendation list for each season, where P1 is the adjusted final recommendation ratio of seasonal products, and P is the preset basic recommendation ratio of seasonal products. Adjust the recommendation timing of the product recommendation list based on the user's seasonal product purchasing habits.

[0044] For example, a user shows a strong preference for warm clothing, such as sweaters and coats, during autumn and winter. By analyzing the user's behavioral data, including browsing records, click records, and purchase records, it is found that the user's seasonal sensitivity S value in autumn is 0.065, and the seasonal sensitivity S value in winter is 0.072. In the preliminary recommendation list, the basic recommendation ratio P of seasonal products in autumn and winter, such as warm clothing, is preset to 0.3, that is, 30% of the recommended content is seasonal products. Since the user's seasonal sensitivity in autumn and winter is high, exceeding the preset sensitivity threshold of 0.05, it is necessary to adjust the recommendation ratio of seasonal products. According to the adjustment formula P1=P×(1+S), the new recommendation ratio is calculated as follows: the recommendation ratio of seasonal products in autumn is 0.3195, and the recommendation ratio of seasonal products in winter is 0.3216. Therefore, the adjusted recommendation ratio is 31.95% for seasonal products in autumn and 32.16% for seasonal products in winter. Based on the user's purchasing habits, the timing of recommendation is adjusted to one month before the start of the season. For example, in early September before the arrival of autumn and early November before the arrival of winter, the platform will focus on pushing recommended information on warm products such as sweaters and coats.

[0045] Step S104, based on the user's purchase record data of the current season and the historical purchase record data of the season, determine whether the user has a purchase behavior deviation, identify the high-frequency product combinations purchased by the user in different seasons, and generate a product recommendation list for subsequent seasons.

[0046] Through the e-commerce platform database, the user's purchase record data of the current season and the historical purchase record data of the season are obtained; the similarity between the user's purchase record data of the current season and the historical purchase record data of the season is calculated through the cosine similarity method, and it is determined whether the similarity is lower than the preset similarity threshold; if the similarity is lower than the preset similarity threshold, it is determined that the user has a purchase behavior deviation; if the user has a purchase behavior deviation, the actual clicks and purchased products of the user in the current season are obtained, the product recommendation list is adjusted, and the user's purchase record data in different seasons are obtained and organized into a transaction data set format, where each transaction represents a purchase record; a support threshold is set, and the sample is collected. The Apriori algorithm is used to mine frequent itemsets, and the high-frequency product combinations purchased by users in different seasons are identified based on the support threshold. The high-frequency product combinations are product combinations that appear at a frequency higher than the preset frequency threshold. Based on the high-frequency product combinations and their support, the FP-Growth algorithm is used to identify the association rules between the high-frequency product combinations, calculate the confidence and lift, and establish a product association database. The association rules and their support, confidence and lift are stored in the product association database, and the association rules of cross-season products are recorded. Based on the high-frequency product combinations purchased in different seasons and the association rules of cross-season products, a recommendation list of products for subsequent seasons is generated.

[0047] Exemplarily, the user's purchase record data for the current spring and the historical purchase record data for the past few springs are obtained through the database of the e-commerce platform. The similarity between the user's current spring purchase record data and the historical purchase record data is calculated by the cosine similarity method. The result shows that the similarity is 0.45, while the preset similarity threshold is 0.6. Since the similarity is lower than the preset threshold, it is determined that the user's purchase behavior has shifted. The actual clicks and purchases of the user in the current season are obtained. If the user actually clicks and purchases a large number of gardening tools and related products in the current spring, which is significantly different from the fitness equipment and outdoor sports products purchased before, the product recommendation list is adjusted to focus on recommending more gardening tools and related products. The user's purchase record data in spring, summer, autumn and winter are obtained, and these data are organized into a transaction data set format, where each transaction represents a purchase record. The support threshold is set to 0.2, and the Apriori algorithm is used for frequent item set mining. It is found that the user's high-frequency purchase of product combinations in spring includes gardening tools and flower pots, with a support of 0.25, and the high-frequency purchase of product combinations in summer includes sunscreen and swimsuits, with a support of 0.3. The FP-Growth algorithm is used to identify association rules between high-frequency product combinations. The calculated confidence and lift are: the confidence of the association rule between gardening tools and flower pots is 0.8, the lift is 1.5, and the confidence of the association rule between sunscreen and swimsuits is 0.85, the lift is 1.7. These association rules and their support, confidence, and lift are stored in the product association database, and the association rules of cross-seasonal products are recorded. Based on the association rules of these cross-seasonal products, a list of recommended products for the subsequent summer is generated. Since the user purchased gardening tools in the spring, and there is a high confidence and lift showing that gardening tools and flower pots are associated, flower pots are added to the summer recommendation list. At the same time, since users have a high habit of buying sunscreen and swimsuits in summer, relevant products are pushed in the summer recommendation list in advance.

[0048] Step S105, based on the commodity sales data, user purchase behavior data and commodity attribute data of different seasons, calculate the relevance score and diversity score of each commodity in the commodity recommendation list in the current season, and balance the seasonal relevance and commodity diversity of the recommended commodities in the commodity recommendation list based on the multi-objective optimization function, determine the best recommended commodity set, and adjust the commodity recommendation list.

[0049] Through the e-commerce platform database, we can obtain the sales data of goods in different seasons, the user purchase behavior data and the attribute data of goods, including price, brand and function. According to the seasonal correlation indicator formula Calculate the relevance score of each product in the product recommendation list in the current season, where R ij is the sales volume of product i in season j, W j is the weight of season j, and N is the total number of seasons. According to the product diversity index formula Calculate the diversity score of the recommended product set in the current product recommendation list, where M is the total number of products and S is the kl is the similarity between product k and product l, which is obtained by using the cosine similarity calculation method based on the attribute data of the products. Balance the seasonal relevance and product diversity of the recommended products in the product recommendation list and determine the best recommended product set, where α and β are weight parameters used to adjust the balance between relevance and diversity. Adjust the product recommendation list based on the best recommended product collection, dynamically update the seasonal relevance and product diversity data, and adjust the product recommendation list based on the real-time calculation results of the multi-objective optimization function F.

[0050] For example, the sales data of goods in different seasons, user purchase behavior data and attribute data of goods are obtained through the e-commerce platform database. To optimize the recommendation system, the seasonal relevance index SRI of goods is calculated. If the sales volume of goods A in winter is 500 pieces, the sales volume of goods B in winter is 300 pieces, and the sales volume of goods C in winter is 200 pieces. If the weight Wj of winter is set to 0.4, the total number of seasons N is 4. Use the seasonal relevance index formula Calculate the SRI of each product and get the winter relevance score SRI of product A as 50, the winter relevance score SRI of product B as 30, and the winter relevance score SRI of product C as 20. ij is the sales volume of product i in season j, W j is the weight of season j, and N is the total number of seasons. There are 3 products in the recommendation list. The attribute data of product A is 5 yuan, brand 7, and function 8. The attribute data of product B is 4 yuan, brand 7, and function 9. The attribute data of product C is 1 yuan, brand 3, and function 2. According to the preset attribute coding, the attribute vector of product A is A = [5,7,8], the attribute vector of product B is B = [4,7,9], and the attribute vector of product C is C = [1,3,2]. Using the cosine similarity calculation method, it is found that the attribute similarity between product A and product B is 0.8, the similarity between product A and product C is 0.2, and the similarity between product B and product C is 0.5. The total number of products M is 3. According to the product diversity index formula Calculate the diversity score DI of the recommended product set in the current product recommendation list, and get DI as 0.3333, where M is the total number of products and S is the total number of products. kl is the similarity between product k and product l. Use multi-objective optimization function Balance seasonal relevance and product diversity in the product recommendation list. If the weight parameters α=1.5 and β=0.5, calculate the value of F and get F=149.8334, which is greater than the preset F value threshold of 120, indicating that the recommendation list maintains sufficient diversity while ensuring seasonal relevance. The larger the F value, the more recommended products are both highly relevant to the current season and not very similar in product category, which improves the recommendation quality and user satisfaction. Therefore, the product combination in the recommendation list has a high relevance score and maintains good diversity, so they will be recommended first in the winter recommendation list. The product recommendation list will be dynamically updated with the real-time calculated F value to ensure that the seasonal relevance and diversity of the recommended products are reasonably balanced. If there are multiple products in the product combination that are too similar and have low diversity scores, some products need to be replaced to increase different types of products, such as adding different categories of winter supplies, such as down jackets, hand warmers, etc. At the same time, according to the latest sales and user behavior data, the seasonal relevance score and diversity score of the product are dynamically updated to optimize the recommendation list in real time. This adjustment ensures that the recommendation list can not only meet users' demand for seasonal products, but also provide diverse choices and enhance users' shopping experience.

[0051] Step S106, monitoring the product recommendation effect in real time, and adjusting the product recommendation list according to the product recommendation effect.

[0052] Obtain real-time product recommendation monitoring data through the e-commerce platform database, compare the recommended products with the products that users actually click on and purchase, and determine the recommendation accuracy. Record the behavior data of users clicking on recommended products and calculate the user click rate. Calculate the purchase conversion rate by comparing the behavior data of users clicking on recommended products with the actual purchase of products. Based on the analysis results of recommendation accuracy, user click rate and purchase conversion rate, as well as the preset product recommendation evaluation standards, including the target values ​​of recommendation accuracy, user click rate and purchase conversion rate, determine whether the product recommendation effect meets expectations. If the product recommendation effect does not meet expectations, adjust the product recommendation list based on the actual clicks and purchases of users.

[0053] Exemplarily, real-time commodity recommendation monitoring data is obtained through the e-commerce platform database to compare the recommended commodities with the commodities actually clicked and purchased by the user, so as to determine the recommendation accuracy. If there are 100 commodities currently recommended to the user, of which the user actually clicked 30 and actually purchased 10, the recommendation accuracy is 30% by calculation. The behavior data of the user clicking on the recommended commodities is recorded, and the user click-through rate is calculated. If the total number of recommended commodity display times is 1,000 times, of which the user clicked 50 times, the click-through rate is 5%. By comparing the behavior data of the user clicking on the recommended commodities with the actual purchase of commodities, the purchase conversion rate is calculated. If the user clicked on the recommended commodities 50 times, of which 10 times were actually purchased, the purchase conversion rate is 20%. According to the analysis results of the recommendation accuracy, user click-through rate and purchase conversion rate, compared with the preset commodity recommendation evaluation standards, if the preset target values ​​are respectively 40% recommendation accuracy, 6% click-through rate and 25% purchase conversion rate. The current recommendation accuracy is 30%, the click-through rate is 5%, and the purchase conversion rate is 20%, all of which have not reached the preset target values. Since the product recommendation effect did not meet expectations, the product recommendation list will be adjusted based on the actual clicks and purchases of users. According to the data of actual clicks and purchases of users, it is found that users are more inclined to buy electronic products and sports equipment, while a large number of products in the recommendation list are household goods and clothing. Therefore, more electronic products and sports equipment are added to the new recommendation list, while the proportion of household goods and clothing is reduced.

[0054] like Figure 2 In this embodiment, a commodity recommendation system based on artificial intelligence may specifically include:

[0055] The seasonal sensitivity analysis module is used to determine whether a product is a seasonal product based on the product's sales time, sales volume, and sales fluctuation indicators, and to calculate the user's seasonal sensitivity based on the user's purchasing behavior data in each season, and to determine the user's seasonal sensitivity type in different seasons.

[0056] The user purchase habit analysis module is used to determine the user's seasonal commodity purchase habits based on the user behavior data of different seasons.

[0057] The product recommendation module is used to determine the user's product preferences through user behavior data, generate a preliminary product recommendation list for each season, adjust the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determine the timing of product recommendations in different seasons based on the user's purchasing habits classification results.

[0058] The cross-season product recommendation module is used to determine whether the user has a purchase behavior deviation based on the user's current season purchase record data and the historical purchase record data of the season, identify the high-frequency product combinations purchased by the user in different seasons, and generate a product recommendation list for subsequent seasons.

[0059] The product recommendation adjustment module is used to calculate the relevance score and diversity score of each product in the product recommendation list in the current season based on the product sales data, user purchase behavior data and product attribute data in different seasons, and balance the seasonal relevance and product diversity of the recommended products in the product recommendation list based on the multi-objective optimization function, determine the best recommended product set, and adjust the product recommendation list.

[0060] The product recommendation effect monitoring module is used to monitor the product recommendation effect in real time and adjust the product recommendation list according to the product recommendation effect.

[0061] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A commodity recommendation method based on artificial intelligence, characterized in that: The method comprises: Determine whether a product is a seasonal product based on its sales time, sales volume, and sales fluctuation index. Calculate the user's seasonal sensitivity based on the user's purchase behavior data in each season and determine the user's seasonal sensitivity type in different seasons. Determine the user's seasonal product purchasing habits based on the user's behavior data in different seasons; Determine the user's product preferences through user behavior data, generate a preliminary product recommendation list for each season, adjust the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determine the timing of product recommendations in different seasons based on the user's purchase habit classification results; Based on the user's purchase record data for the current season and the historical purchase record data for the season, determine whether the user has a purchase behavior deviation, identify the high-frequency product combinations purchased by the user in different seasons, and generate a product recommendation list for subsequent seasons; Based on the sales data of goods in different seasons, user purchase behavior data and attribute data of goods, the relevance score and diversity score of each product in the product recommendation list in the current season are calculated. Based on the multi-objective optimization function, the seasonal relevance and product diversity of the recommended goods in the product recommendation list are balanced, the best recommended product set is determined, and the product recommendation list is adjusted. Monitor the product recommendation effect in real time and adjust the product recommendation list based on the product recommendation effect.

2. The method according to claim 1, wherein: The method of judging whether a product is a seasonal product based on the sales time, sales volume and sales fluctuation index of the product, and calculating the seasonal sensitivity of the user based on the purchase behavior data of the user in each season, and judging the seasonal sensitivity type of the user in different seasons, includes: The sales time and sales volume of the goods are extracted from the e-commerce platform database, and the obtained data is cleaned to remove outliers and missing values, and the date field is converted into a time series format; the moving average method is used to calculate sales fluctuations to obtain sales fluctuation indicators of the goods, including sales peak and valley values, standard deviation and variance of sales, and seasonal index. The seasonal index is the ratio of the sales volume of the goods in different seasons to the average sales volume; based on the sales volume and sales fluctuation indicators of the goods, the K-means clustering algorithm is used for model training to determine whether the goods are seasonal goods; the purchase record data of seasonal goods purchased by users in different seasons are extracted from the sales records in the e-commerce platform database. The purchase record data includes product ID, product category, purchase time, purchase quantity and purchase amount; the user's seasonal product purchase record data is divided by season to determine the user's purchase behavior data in each season. The purchase behavior data includes purchase frequency, purchase volume and purchase amount; based on the user's purchase behavior data in each season, the seasonal sensitivity evaluation formula is used Calculate the seasonal sensitivity of the user, where F represents the purchase frequency of seasonal products, Q represents the purchase volume of seasonal products, A represents the purchase amount of seasonal products, and F represents the purchase frequency of seasonal products. Total represents the total purchase frequency of all commodities in the season, Q Total A represents the total purchase volume of all commodities in this season. Total represents the total purchase amount of all commodities in the season, S represents the seasonal sensitivity of the user, χ represents the influence of purchase frequency, δ represents the influence of purchase volume, ε represents the influence of purchase amount, and φ represents the index adjustment factor, which is used to control the overall influence of purchase frequency, purchase volume and purchase amount when calculating seasonal sensitivity. χ, δ, ε and φ are all obtained by exploratory analysis of historical purchase data. According to the calculated seasonal sensitivity, the sensitivity threshold is set to determine the seasonal sensitivity type of the user in different seasons, including high seasonal sensitivity and low seasonal sensitivity.

3. The method according to claim 1, wherein: Determining the user's seasonal commodity purchasing habits based on the user's behavior data in different seasons includes: Through the e-commerce platform database, query the user behavior data of users in different seasons based on user ID and mark the user's seasonal sensitivity type. User behavior data includes browsing history, click history and purchase history data; According to the preset time period, the user behavior data of users in different seasons are divided into different time intervals, and the user behavior data are further divided according to the classification labels of the goods, and the classification labels of the goods include seasonal goods and non-seasonal goods; the user behavior data are grouped according to user characteristics, and the user characteristics include age, gender, and geographical location; according to the purchase frequency, purchase quantity and purchase amount of different commodity categories of each user in each time period of different seasons, the decision tree algorithm is used to train the model to determine the user's seasonal commodity purchasing habits, which include advance purchase, immediate purchase and delayed purchase.

4. The method according to claim 1, wherein: The method of determining the user's product preferences through user behavior data, generating a preliminary product recommendation list for each season, adjusting the recommendation ratio of seasonal products in the preliminary recommendation list based on the user's seasonal sensitivity in each season, and determining the product recommendation timing in different seasons in combination with the user's purchase habit classification results, includes: Determine the user's product preferences based on user behavior data; According to the user's product preferences in each season, a product-based collaborative filtering recommendation algorithm is used to calculate the similarity between products, and a preliminary product recommendation list for each season is generated using the similarity matrix; The seasonal sensitivity evaluation formula is used to obtain the seasonal sensitivity of the user in each season through the user's purchase behavior data in each season; if the user is a high seasonal sensitivity user, the seasonal product recommendation ratio adjustment formula P1=P×(1+S) is used according to the user's seasonal sensitivity to determine the final recommendation ratio of seasonal products, adjust the seasonal product recommendation ratio in the preliminary recommendation list, and generate a product recommendation list for each season, where P1 is the adjusted final recommendation ratio of seasonal products, and P is the preset basic recommendation ratio of seasonal products; Adjust the timing of product recommendation lists based on users' seasonal product purchasing habits.

5. The method according to claim 1, wherein: The method of judging whether the user has a purchase behavior deviation based on the user's purchase record data of the current season and the historical purchase record data of the season, identifying the high-frequency product combinations purchased by the user in different seasons, and generating a product recommendation list for subsequent seasons includes: Obtain the user's purchase record data of the current season and the historical purchase record data of the season through the e-commerce platform database; calculate the similarity between the user's purchase record data of the current season and the historical purchase record data of the season through the cosine similarity method, and determine whether the similarity is lower than a preset similarity threshold; if the similarity is lower than the preset similarity threshold, determine that the user has a purchase behavior deviation; If the user has a purchase behavior deviation, obtain the actual clicks and purchased products of the user in the current season, adjust the product recommendation list, and obtain the user's purchase record data in different seasons and organize it into a transaction data set format. Each transaction represents a purchase record. A support threshold is set, and the Apriori algorithm is used to mine frequent itemsets. The high-frequency commodity combinations purchased by users in different seasons are identified based on the support threshold. The high-frequency commodity combinations are commodity combinations that appear at a frequency higher than the preset frequency threshold. Based on the high-frequency commodity combinations and their support, the FP-Growth algorithm is used to identify the association rules between high-frequency commodity combinations, calculate the confidence and lift, and establish a commodity association database. The association rules and their support, confidence and lift are stored in the commodity association database, and the association rules of cross-season commodities are recorded. Based on the high-frequency commodity combinations purchased in different seasons and the association rules of cross-season commodities, a recommendation list of commodities for subsequent seasons is generated.

6. The method according to claim 1, wherein: The method calculates the relevance score and diversity score of each product in the product recommendation list in the current season based on the product sales data, user purchase behavior data and product attribute data of different seasons, and balances the seasonal relevance and product diversity of the recommended products in the product recommendation list based on the multi-objective optimization function, determines the best recommended product set, and adjusts the product recommendation list, including: Through the e-commerce platform database, obtain the sales data of goods in different seasons, user purchase behavior data and attribute data of goods, including price, brand and function; according to the seasonal correlation indicator formula Calculate the relevance score of each product in the product recommendation list in the current season, where R ij is the sales volume of product i in season j, W j is the weight of season j, N is the total number of seasons; according to the product diversity index formula Calculate the diversity score of the recommended product set in the current product recommendation list, where M is the total number of products and S is the kl is the similarity between product k and product l, which is obtained by using the cosine similarity calculation method based on the attribute data of the products; according to the multi-objective optimization function Balance the seasonal relevance and product diversity of the recommended products in the product recommendation list, and determine the best recommended product set, where α and β are weight parameters used to adjust the balance between relevance and diversity; adjust the product recommendation list according to the best recommended product collection, and dynamically update the seasonal relevance and product diversity data, and adjust the product recommendation list based on the real-time calculation results of the multi-objective optimization function F.

7. The method according to claim 1, wherein: The real-time monitoring of the product recommendation effect and adjusting the product recommendation list according to the product recommendation effect include: Obtain real-time product recommendation monitoring data through the e-commerce platform database, compare the recommended products with the products actually clicked and purchased by users, and determine the accuracy of the recommendations; Record the behavior data of users clicking on recommended products and calculate the user click rate; calculate the purchase conversion rate by comparing the behavior data of users clicking on recommended products with the actual purchase of products; judge whether the product recommendation effect meets expectations based on the analysis results of recommendation accuracy, user click rate and purchase conversion rate, as well as the preset product recommendation evaluation standards, including the target values ​​of recommendation accuracy, user click rate and purchase conversion rate; If the product recommendation effect does not meet expectations, the product recommendation list will be adjusted based on the user's actual clicks and purchases.

8. A commodity recommendation system based on artificial intelligence, characterized in that: The system comprises: Seasonal sensitivity analysis module, used to determine the seasonal sensitivity type of users in different seasons based on the purchasing behavior data of users in different seasons; User purchasing habit analysis module, used to determine the user's seasonal commodity purchasing habits based on the user's behavior data in different seasons; The product recommendation module is used to determine the user's product preferences through user behavior data, generate a preliminary product recommendation list for each season, and adjust the recommendation ratio of seasonal products in the preliminary recommendation list; The cross-seasonal product recommendation module is used to determine whether the user has a purchase behavior deviation, identify the high-frequency product combinations purchased by users in different seasons, and generate a product recommendation list for subsequent seasons; The product recommendation adjustment module is used to calculate the relevance score and diversity score of each product in the product recommendation list in the current season, determine the best recommended product set, and adjust the product recommendation list; The product recommendation effect monitoring module is used to monitor the product recommendation effect in real time and adjust the product recommendation list according to the product recommendation effect.

Images