Intelligent order recommendation method and electronic equipment

Abstract

The invention discloses an order intelligent recommendation method and electronic equipment, and the method comprises the steps: obtaining a scene tag in response to a recommendation request, and searching a target commodity variety according to the scene tag; obtaining a user tag according to the recommendation request; obtaining a pre-configured real-time recommendation tag; and matching a target commodity in the target commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity. According to the method, the intelligent recommendation process is performed by integrating the current consumption scene, the characteristics of the user and the demands of the merchant commodities, the matching degree between the recommended target commodities and the demands of the user and the merchant is improved, and the efficiency of screening the target commodities by the user is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of prediction, in particular to an order intelligent recommendation method and electronic equipment. BACKGROUND

[0002] When having a purchase demand, a user usually expects a platform to recommend products meeting his / her expectation. The existing recommendation algorithm of the platform is basically based on the advertisement of a merchant and the user portrait for comprehensive recommendation, which is poor in real-time performance and cannot accurately recommend the demand of a user generated temporarily.

[0003] For example, in a KTV room scenario, a user needs to select desired goods in a short time. The user needs to spend a lot of time in browsing and searching and then selecting, and direct provision of a package cannot meet the personalized demand. SUMMARY

[0004] The present application aims to solve the technical problem of providing an order intelligent recommendation method and electronic equipment to improve the efficiency of recommended goods.

[0005] To solve the above technical problem, the technical solution adopted by the present application is as follows: An order intelligent recommendation method comprises the following steps: In response to a recommendation request, a scene label is acquired, and a target product variety is found according to the scene label; A user label is acquired according to the recommendation request; A pre-configured real-time recommendation label is acquired; A target product is matched in the target product variety according to the user label and the real-time recommendation label, and the target product is recommended.

[0006] To solve the above technical problem, another technical solution adopted by the present application is as follows: An electronic equipment comprises a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements each step of the above order intelligent recommendation method when executing the computer program.

[0007] The present application has the beneficial effect that when recommending goods, the target product variety associated with the scene label is acquired according to the current scene label, the user label and the real-time recommendation label are acquired, and the target product is matched in the target product variety according to the user label and the real-time recommendation label within the range of the target product variety and is recommended. In this way, the current consumption scene, the characteristics of the user and the demand of the merchant goods are comprehensively considered in the intelligent recommendation process, the matching degree of the recommended target product with the demand of the user and the merchant is improved, and the efficiency of the user in selecting the target product is improved. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A step flow chart in an order intelligent recommendation method provided for an embodiment of the present application; Figure 2 A component structure schematic diagram of an electronic device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0009] To explain the technical content, the achieved purposes and effects of the present application in detail, the following will be explained in combination with embodiments and the accompanying drawings.

[0010] In the related art, users have the demand of quickly screening out suitable goods in some special purchase scenarios. For example, in a KTV, it is usually a multi-person gathering scenario, and users expect to spend time in communication with others. The goods provided in the KTV are in large quantity and are usually only simply classified and displayed. Users need to browse, search and screen according to the demand, which consumes a lot of time and energy. Moreover, if the user is not familiar with the KTV scenario, he or she will not know which target goods are suitable for the current gathering scenario, such as company team building scenario, friend gathering scenario and birthday scenario, and the corresponding goods demand will be different. Moreover, for the scenario of selecting target goods according to the total budget, the user can only adjust the purchase strategy according to the current total price of goods while checking, which further reduces the purchase efficiency.

[0011] Please refer to Figure 1 , which discloses an order intelligent recommendation method, comprising steps 110 to 140.

[0012] Step 110, acquiring a scene label in response to a recommendation request, and finding a target goods variety according to the scene label. The scene label can be obtained based on current room data analysis, or can be a scene label selected by the user. For example, if a happy birthday song is read in the current song list, it can be obtained that the current scene label is a birthday scene. For example, a preset scene label is displayed through a front-end interactive layer, such as a birthday scene, a business scene, a general gathering scene, etc. The user can click the scene label to select. The expected order result can also be input, and the order result is segmented and high-frequency labels are counted by the background, which are recorded as the user's common labels for subsequent direct display.

[0013] Step 120, acquiring a user label according to the recommendation request.

[0014] Step 130, acquiring a pre-configured real-time recommendation label.

[0015] Step 140, matching target goods in the target commodity variety according to the user label and the real-time recommendation label and recommending the target goods. The way to match the target goods according to the user label and the real-time recommendation label can be to input the user label, the real-time recommendation label and the target commodity variety into a rule engine, and output the corresponding target goods by the rule engine. The rule engine can be configured based on Drools, QLExpress, etc. For example, ① configure the scene label for the front end to show for the consumer to click, such as “birthday party” and “company team building”; ② configure the label core rule: divided into “mandatory commodity variety”, “preferred commodity variety” and “excluded commodity variety”, each scene label is associated with mandatory commodity variety, preferred commodity variety and excluded commodity variety, such as “birthday party”, the mandatory commodity variety is configured as “cake, champagne, sparkling wine, fruit plate” for or relationship, the preferred commodity variety is configured as “decoration, longevity face, beer” for or relationship, and the “excluded commodity variety” is configured as “cigarette”. The core rule in the rule engine can be uniformly configured by the background administrator, or can be generated by AI, for example, after configuring the scene label, let AI output the core rule according to the format of [scene name]->mandatory commodity variety [X], preferred commodity variety [Y] and excluded commodity variety [Z].

[0016] The user label can be obtained according to the user's historical order data, and a dynamic interest vector is generated for each user: combining the user's historical selection of commodity features, the preference weight is calculated. For example, the labels corresponding to the commodity categories in the user's order can be obtained, the labels with a proportion greater than a preset proportion are counted, for example, the labels with a proportion greater than 60% are counted, and the weight is obtained according to the proportion, which can be directly used as the weight; for example, if 80% of the user's orders contain the “low-price snack” label, the user label includes “low-price snack”, and the weight of low-price goods is set to 0.8 in the process of matching the target goods this time, and 60% of the user's orders contain fruit plate, so the user label includes “fruit plate” and the corresponding weight is set to 0.6.

[0017] Real-time factors are considered, and sometimes merchants need to set some commodity weights according to popular commodities, inventory status and actual demand, such as XX wine aluminum can goods, which are in short supply and have more inventory, and are expected to be preferentially recommended for sale. Therefore, the name of the goods can be directly used as the real-time recommendation label, and the weight of the goods in the subsequent recommendation is divided by 0.8.

[0018] From the above description, the beneficial effects of the present application are that: in the process of commodity recommendation, first, the target commodity variety associated with the scene label is obtained according to the current scene label, then the user label and the real-time recommendation label are obtained, and the target commodity is matched in the target commodity variety according to the user label and the real-time recommendation label within the range of the target commodity variety and is recommended, so that the current consumption scene, the characteristics of the user himself and the demand of the merchant commodity are comprehensively considered in the intelligent recommendation process, the matching degree of the recommended target commodity with the demand of the user and the merchant is improved, and the efficiency of the user in screening the target commodity is improved.

[0019] In an embodiment of the present application, before the target commodity variety is found according to the scene label in step 110, steps 101 to 103 are included.

[0020] Step 101, obtaining historical order data, which includes product varieties and scene tags. The tags corresponding to product varieties can also include budget tags, price intervals, product categories, etc., to allow users to make refined selections. There are two ways to tag product varieties: one is to record the name, specifications, and taste of the product variety for non-standardized product varieties, such as platter and snack; the other is to tag according to historical order data, for example: a) When a product variety is normally sold online, the orders it is in are often in high-value historical orders, i.e., orders with a total amount exceeding a preset amount, such as orders exceeding 500 yuan, and the product variety's budget tag will be marked as high-value, and the price interval tag will be the average price interval of the historical orders in which the product variety is located, such as 500-600 yuan; b) A product variety is often sold in a specific merchant type, and will be tagged with a merchant-related tag, such as "party" or "bulk" in a party KTV or bulk KTV, and will be tagged with "party", "bulk", "business", etc. scene tags; c) A product is often sold in a specific box type, and the orders in a specific box type account for a preset proportion of the total number of orders, and will be tagged with the corresponding box tag, such as A product 70% of the orders come from large boxes, and the data cleaning automatically tags it with the "large box" tag, and if the product specifications are large capacity, i.e., the weight or volume of the product exceeds a preset value, such as 1L, large plate, etc., it will be supplemented with the "share with many people" tag; d) Tagging based on sales time period, if the number of orders of A product in a special time period exceeds a preset proportion of the total number of orders, and the duration exceeds a preset time, it will be tagged with a special time period tag, such as early stage, midnight stage, and if the sales of a product in a specific time, such as weekends, Valentine's Day, World Cup, etc. exceed a preset proportion, such as 60% higher than usual, it will automatically add "weekend", "holiday", "World Cup", etc. tags; e) Tagging based on recommended successful orders, the target product of each recommendation will be recorded in detail in the database, and the background will analyze the product variety and scene tag in the successful orders every day, such as A product is often ordered in the "birthday party" scene tag, and will be tagged with the "birthday party" scene tag.

[0021] Step 102, calculating the adaptation value of each product variety and the scene tag. For example, "whisky" has an adaptation degree of 0.76 in the "business reception" scene tag.

[0022] Step 103, obtaining the required product variety and preferred product variety corresponding to each scene tag according to the adaptation value.

[0023] Step 110, searching for target product varieties according to the scene tags includes step 111: searching for the required product variety and preferred product variety corresponding to the scene tag as target product varieties according to the scene tag.

[0024] As can be seen from the above description, the product variety and the scene tag in the historical order data of the Palace Museum are used to establish the association between the product variety and the scene tag, so that when the scene tag corresponding to the recommendation request is received, the mandatory product variety and the preferred product variety corresponding to the scene tag can be directly obtained, and the target product is determined within the range of the mandatory product variety and the preferred product variety, thereby improving the accuracy of the selection of the target product by taking the historical order data as the recommendation reference, and associating the selection of the target product with the scene tag, and the scene tag corresponds to the respective target product variety, so as to avoid the deviation of the recommendation result from the actual demand.

[0025] In an embodiment of the present application, step 210 of obtaining the excluded product variety corresponding to the scene tag is further included.

[0026] Step 140 further includes step 141 of obtaining the recommended product by eliminating the excluded product variety from the target product, and recommending the recommended product.

[0027] As can be seen from the above description, for each scene tag, the excluded product variety is also set, and after obtaining the target product, the excluded product variety is checked, the excluded product variety is eliminated from the target product to obtain the recommended product, and the recommended product is recommended, so as to avoid the occurrence of the excluded product variety that does not match the scene in the final recommended product to the user, and to ensure the adaptation of the recommendation result to the scene tag. Moreover, the recommended product is a further screening of the target product, and therefore the further processing of the target product can also be applied to the processing of the recommended product.

[0028] In an embodiment of the present application, step 102 includes F = n / N; in the formula, F represents the adaptation value, n represents the number of times the product variety is purchased under the scene tag, and N represents the total number of times the product variety is purchased.

[0029] As can be seen from the above description, by taking the ratio of the number of times the product variety is purchased under a scene tag to the total number of times the product variety is purchased as the adaptation value of the product variety to the scene, the product variety strongly associated with the scene can be found to directly perform the recommendation to reduce the time of user screening and improve the recommendation accuracy of the target product; and the adaptation value of a product variety under each scene tag can also be obtained, if the difference between all adaptation values of a product is within a preset range, the product is marked as a general product variety, and after the target product variety is found in step 110, the general product variety is also added to the target product variety to perform the subsequent steps, so as to avoid the problem that the general product is missed in the recommendation in the case that there are many scene tags, because the general product is purchased in the order corresponding to each scene tag, so that the adaptation value corresponding to each scene is not high.

[0030] In one embodiment of the present application, the method further comprises step 310: receiving a budget amount.

[0031] Step 140 further comprises steps 142-144.

[0032] Step 142, obtaining a unit price of the target commodity.

[0033] Step 143, calculating the quantity of the target commodity according to the budget amount and the unit price. Wherein, the quantity of each target commodity can be initially set to 1, and the current total price is calculated. If the ratio of the current total price to the budget amount is within a preset range, the quantity of the target commodity is directly returned to 1. Otherwise, the weights of all target commodities are obtained, and the quantity of the target commodity is increased one by one according to the order from high to low of the weights until the ratio of the current total price to the budget amount is within the preset range. The weight is the weight matched according to the user label or the real-time recommended label in the above steps; for example, the target commodities include beer and chips, the current total price is 10 when the quantity of beer is 1 and the quantity of chips is 1, the budget amount is 26, the weight of beer is 0.9 and the unit price is 6, the weight of chips is 0.7 and the unit price is 4, and the preset range is 0.8-1.2; 10 / 17 is not within the preset range, the quantity of beer is increased by 1 according to the weight, the current total price is 16, 16 / 26 is not within the preset range, then the quantity of chips is increased by 1, the current total price is 20, 20 / 26 is not within the preset range, then the quantity of beer is increased by 1, the current total price is 26, 26 / 26 is within the preset range, then the process is stopped and the current target commodities and the quantity of the target commodities are output: beer 3, chips 2.

[0034] Step 144, returning the target commodities and the quantity of the target commodities.

[0035] The budget amount can be a directly input amount by the user, or a threshold range of the budget amount can be configured through a backend configuration page as a budget label, such as 10% indicating that the recommended result of a 1000 yuan budget can be between 900-1100 yuan. After the above configuration is updated, the backend recommendation rule engine will automatically parse this rule and process the recommended result according to the rule. For example: “friend's birthday” -> the selected commodity varieties include cakes, champagne / sparkling wine, and birthday theme decorations. “1000 yuan budget” -> the total price of the commodities needs to be close to 1000 yuan (such as 950-1050 yuan). “Company team building” -> the preferred commodity varieties include beer, snack platters, and fruit platters, and the excluded commodity varieties include high-priced beverages, such as beverages with a unit price exceeding 50 yuan.

[0036] From the above description, the user can directly input the budget amount, if there is a corresponding budget amount, after obtaining the target commodity, the corresponding commodity unit price of the target commodity is also obtained, the quantity of the target commodity is determined according to the budget amount and the commodity unit price, and the target commodity and the quantity of the target commodity are returned together, which further reduces the process of determining the quantity of the target commodity by the user. The user only needs to simply add or subtract the quantity on the basis of the recommendation result, thereby improving the ordering efficiency.

[0037] In an embodiment of the present application, step 410 is further included: calculating the correlation degree between each two commodity varieties, and if the correlation degree exceeds a correlation degree threshold, then the two commodity varieties are associated and stored in an association table.

[0038] Step 140 further includes steps 145 to 146.

[0039] Step 145, obtaining the associated commodity variety corresponding to the target commodity variety from the association table.

[0040] Step 146, matching the target commodity in the target commodity variety and the associated commodity variety according to the user label and the real-time recommendation label and recommending the target commodity.

[0041] From the above description, the correlation degree between two commodity varieties is calculated in advance, and when the correlation degree exceeds a correlation degree threshold, the two commodity varieties are associated and stored in an association table. Therefore, in the association table, another commodity variety associated with a determined commodity variety can be found. After determining the target commodity, the associated commodity variety corresponding to the target commodity variety is found, and subsequent operations are performed based on the target commodity variety and the associated commodity variety. On the basis of matching the target commodity variety based on the scene label, the range of finally determining the target commodity is widened, and the possibility of obtaining a target commodity empty result after introducing the user label and the real-time recommendation label is reduced, thereby improving the accuracy of the recommended target commodity.

[0042] In an embodiment of the present application, the calculation of the correlation degree between each two commodity varieties in step 410 includes: G=m / max(Xa, Xb); wherein G represents the correlation degree, max(Xa, Xb) represents the maximum value of Xa and Xb; Xa represents the number of times that commodity variety a appears in historical orders, Xb represents the number of times that commodity variety b appears in the historical orders, and m represents the number of historical orders in which commodity variety a and commodity variety b appear together.

[0043] From the above description, if the proportion of two commodities appearing in an order at the same time is large, it indicates that the user tends to purchase the two commodities at the same time, and therefore, determining the relevance between two commodity varieties by the ratio of the number of orders in which the two commodity varieties appear at the same time to the total number of orders in which the commodity varieties appear can accurately determine whether the two commodity varieties are relevant, so that when the target commodity variety has relevant commodity varieties, the relevant commodity varieties are also recommended, so as to improve the richness of the target commodity and improve the matching degree of the recommendation result and the user's expectation.

[0044] In an embodiment of the present application, steps 510 to 530 are further included.

[0045] Step 510, converting the commodity varieties into commodity vector representations.

[0046] Step 520, calculating the similarity between each two commodity vector representations by a cosine similarity algorithm.

[0047] Step 530, for each commodity variety, recording the top pre-set number of similar commodity varieties with the highest similarity to the commodity variety to obtain a similarity table. For example, matching the top 10 commodities with the highest similarity to obtain the similarity table corresponding to the commodity variety; for example, the similarity table of “imported beer” includes similar commodity varieties “potato chips” (similarity 0.72) and “nut platter” (similarity 0.68), and the similar commodity varieties in the similarity table are sorted in descending order of similarity.

[0048] Step 140 further includes steps 147 to 148.

[0049] Step 147, obtaining the similar commodity variety corresponding to the target commodity from the similarity table.

[0050] Step 148, matching the target commodity and recommending the target commodity in the target commodity variety and the similar commodity variety according to the user label and the recommendation label.

[0051] It can be seen that the similarity between the two commodity varieties is calculated, specifically by converting the commodity varieties into commodity vector representations, calculating the similarity between each two commodity vector representations through a cosine similarity algorithm, saving, for each commodity variety, the commodity varieties corresponding to the ten commodity vector representations with the highest similarity to the commodity vector representation of the commodity variety; it can be seen that the similarity between the commodity vector representations is the similarity between the commodity varieties, and a similarity table recording the similar commodity varieties of each commodity variety is obtained; after obtaining the target commodity variety, the similar commodity varieties corresponding to the target commodity variety are obtained from the similarity table, and subsequent steps are performed based on the target commodity variety and the similar commodity varieties, so as to expand the basis of screening and improve the richness of the final recommended target commodity.

[0052] In an optional real-time manner, before step 145 or step 147, a data preprocessing process is further included: structuring the historical order data to construct a “user-commodity variety-scene label” three-dimensional matrix, including user ID (Identity document, identity), commodity ID, scene label (such as “birthday party” “company team building”), purchase frequency, click frequency and other characteristics. For example, ①mark the commodities clicked by the user but not purchased as “potential interest” labels, and assign a basic weight of 0.3 ②mark the high-frequency repeat purchase commodity combinations in the recent preset number of days (such as “cake + champagne” high-frequency repeat purchase in the last 90 days) as “strongly associated combinations”, which can also be directly recorded in the temporary association table, because it is updated with the recent time.

[0053] It can be seen that one or more of the above steps 210, 410 and 510 can be selected for execution, or all of them can be executed; if all of them are executed, after obtaining the target commodity variety, the associated commodity varieties are found from the association table and the similar commodity varieties are found from the similarity table, and then the target commodity is matched based on the user label and the real-time recommendation label in the range of the target commodity variety, the associated commodity variety and the similar commodity variety, and finally the excluded commodity varieties are excluded from the target commodity to obtain the recommended commodity and perform the recommendation. It can be seen that the subsequent steps can also be performed after obtaining the commodity unit price of the recommended commodity in step 142. For example, when the user selects “beer”, the system calls the similar commodity variety “sparkling wine”, combines the optional commodity variety and the preferred commodity variety under the current scene label (such as the preferred commodity variety under the “company team building” scene label including large packages of snacks and large fruit plates) to obtain the target commodities “beer”, “sparkling wine”, “500g potato chips” and “fruit plate-large”. It can be seen that the commodity variety can be a category of goods, such as “wine” and “snacks”, and the target commodity can be a commodity variety or a specific commodity, such as “beer” or “C brand 250ml beer”.

[0054] The matching process of step 140 includes: 1. According to the "mandatory product variety", "preferred product variety" and "excluded product variety" corresponding to the scene label in the rule engine, the "mandatory product variety" and "preferred product variety" that meet the scene are screened out; according to the content recommendation algorithm, the "mandatory product variety" with high weight (adaptation degree higher than or equal to a first preset value) is selected; according to the collaborative filtering rule, the preferred product variety with high weight (adaptation degree less than the first preset value and greater than or equal to a second preset value) is selected, wherein the first preset value is greater than the second preset value; 2. Any selection process excludes "excluded product variety"; 3. Finally, ensure that the output is a recommended combination of 3-5 products. For example: the user selects a budget of 500, and the scene label is company team building. According to the "mandatory product variety", "preferred product variety" and "excluded product variety" set for company team building, "high-priced wine" is excluded, the "mandatory product variety" beer category is preferentially selected, the highest recommendation weight xxx beer is obtained according to the weight of each beer product obtained by the content recommendation algorithm. According to the weight of the "preferred category" 0.6, the product weight (adaptation degree) of the co-occurrence matrix xxx beer obtained by the collaborative filtering algorithm, if a product is both a "preferred category" and a product with high adaptation degree, the weight will be cumulative, the highest weight xxx snack platter and the second highest weight large fruit plate are selected. And adjust the number of each product according to the 500 budget, limit the number of products in each category, such as 50 bottles of beer at most, two fruit plates at most, etc.

[0055] As can be known from the above description, after obtaining the target product variety, the selection range of the target product is widened by comprehensively associating the product variety and the similar product variety, and a richer selection basis is provided to avoid missing the product expected by the user; the associated product variety is counted from the user's purchase in the historical order, and the similar product variety similar to the product variety is obtained according to the product vector representation of the product variety, other product varieties with a high probability of being recommended together with the target product variety are obtained from different dimensions, and the accuracy of the final target product is improved.

[0056] In an embodiment of the present application, steps 610 to 650 are further included.

[0057] Step 610, a preset large language model template is obtained, and the user label, real-time recommendation label and scene label are filled into the large language model template to obtain a large language model prompt. The large language model (LLM) can be deepseek, chatGPT, etc. The historical order data and rules in the rule engine can be used as learning corpus of the large language model to make the large language model obtain a target commodity recommendation more suitable for the scene in the KTV. The large language model template can be a template with key prompt words left out, for example, "the current scene is

scene label

rule 1

rule 2

user label

real-time recommendation label

budget amount

[0058] Step 620, inputting the large language model prompt into the large language model to obtain a first target commodity.

[0059] Step 630, according to the user label and the real-time recommendation label, a second target commodity is matched in the target commodity variety according to a preset rule.

[0060] Step 640, a first adaptation value of the first target commodity and a second adaptation value of the second target commodity are calculated.

[0061] The specific calculation method of the adaptation value can refer to Table 1 below, and the first adaptation value and the second adaptation value are obtained by weighting five dimensions of rule matching degree, user preference fit degree, commodity synergy score, price rationality and novelty. The label rule that meets the label rule means that the label of the target commodity meets one of the user label, the scene label or the real-time recommendation label; the sour taste of meeting the label rule can also be set to meet two or three according to the strictness, which is not limited here. The budget deviation rate = | (the total price of all target commodities - budget amount) | ÷ budget amount, and the threshold value can be set according to the demand, for example, the allowed deviation is 5% or 10%.

[0062] Table 1

[0063] Step 650, recommending the first target commodity or the second target commodity according to the comparison result of the first adaptation value and the second adaptation value. When recommending the target commodity and the quantity corresponding to the target commodity to the user, the corresponding "+" and "-" buttons can be provided on the page to facilitate the user to adjust the order quantity by himself, and after the user confirms the recommendation page, the user is directly jumped to the order page or is added to the shopping cart to wait for recommendation. It can be seen that if the user finally confirms the recommendation page, the user's confirmation result can be recorded as a historical order, and the target commodity corresponding to the user tag and the scene tag is updated in real time every interval of a preset time period or when the historical order is updated, so as to improve the accuracy of subsequent recommendation.

[0064] As can be known from the above description, after introducing a large language model to obtain the first target commodity, and matching the second target commodity from the target commodity variety according to the preset rule, and obtaining different target commodity recommendation results by different ways, the first adaptation value and the second adaptation value are calculated respectively, and the target commodity is recommended according to the adaptation value. In this way, a more flexible and higher expansion program recommendation mode of the large language model is provided, and the minimum guarantee of the recommendation mode according to the preset rule can be realized to ensure that the target commodity basically matched with the user's expectation can be finally given. By calculating the adaptation value, the target commodity obtained by the two is selected, and the optimal recommendation mode is automatically adapted according to the difference of each recommendation request.

[0065] The application further provides an electronic device 700, comprising a memory 702, a processor 701, and a computer program stored in the memory 702 and running on the processor 701, wherein the processor 701 implements each step of the above-mentioned order intelligent recommendation method when executing the computer program.

[0066] The electronic device 700 in the application can achieve the beneficial effects as described above, which will not be repeated here.

[0067] To sum up, by means of the order intelligent recommendation method and the electronic device provided in the application, the target commodity variety is matched out through the scene label first, and then the target commodity is matched in the target commodity variety based on the obtained user label and the real-time recommendation label, so that the user can obtain the recommended target commodity without understanding the specific commodity, the consideration range of the user is reduced, and the ordering efficiency is improved; further, the user label can be a label obtained from the historical order corresponding to the user, or a label selected by the user in real time, the user changes from selecting a specific commodity to selecting a label, the ordering process of the user is simplified, the amount of data browsed by the user is reduced, and the matching degree between the finally recommended target commodity and the user expectation is improved, that is, the accuracy of the finally recommended target commodity is improved; meanwhile, the new order generated by the user will also be analyzed as the historical order when the user orders next time, so that the automatic optimization process is realized. Furthermore, the form of the large language model is introduced, the interpretation ability of the user intention is improved, and the final target commodity recommended to the user is obtained by calculating the adaptation values of the two different target commodity selection methods, so as to avoid the situation that the flexibility of the large language model is too high and the result deviates.

[0068] The above is only an embodiment of the application, and does not limit the patent scope of the application, and any equivalent transformation or direct or indirect application in the related technical field based on the content of the specification and drawings of the application is also included in the patent protection scope of the application.

Claims

1. An order intelligent recommendation method, characterized in that, The method comprises the following steps: acquiring a scene tag in response to a recommendation request, and searching for a target commodity variety according to the scene tag; acquiring a user tag according to the recommendation request; acquiring a pre-configured real-time recommendation tag; matching a target commodity in the target commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity. 2.The method of claim 1, wherein, Before searching for a target commodity variety according to the scene tag, the method comprises the following steps: acquiring historical order data, wherein the historical order data comprises a commodity variety and a scene tag of each historical order; calculating an adaptation value of each commodity variety and the scene tag; obtaining a mandatory commodity variety and a preferred commodity variety corresponding to each scene tag according to the adaptation value; the step of searching for a target commodity variety according to the scene tag comprises the following steps: searching for the mandatory commodity variety and the preferred commodity variety corresponding to the scene tag as the target commodity variety according to the scene tag. 3.The method of claim 1, wherein, The method further comprises the following step: obtaining an excluded commodity variety corresponding to the scene tag; the step of matching a target commodity in the target commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity comprises the following steps: 4.The method of claim 2, wherein, obtaining a recommended commodity by excluding the excluded commodity variety from the target commodity, and recommending the recommended commodity. the step of calculating an adaptation value of each commodity variety and the scene tag comprises the following steps: F = n / N; 5. The method of claim 1, wherein, in the formula, F represents the adaptation value, n represents the number of times the commodity variety is purchased under the scene tag, and N represents the total number of times the commodity variety is purchased. The method further comprises the following steps: receiving a budget amount; the step of matching a target commodity in the target commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity comprises the following steps: obtaining a commodity unit price of the target commodity; calculating the quantity of the target commodity according to the budget amount and the commodity unit price; 6. The method of claim 1, wherein, returning the target commodity and the quantity of the target commodity. The method further comprises the following steps: calculating the correlation degree between each two commodity varieties, and if the correlation degree exceeds a correlation degree threshold, then storing the two commodity varieties in an association table; the step of matching a target commodity in the target commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity further comprises the following steps: obtaining an associated commodity variety corresponding to the target commodity variety from the association table; 7. The method of claim 6, wherein, matching a target commodity in the target commodity variety and the associated commodity variety according to the user tag and the real-time recommendation tag, and recommending the target commodity. the step of calculating the correlation degree between each two commodity varieties comprises the following steps: G = m / max (Xa, Xb); 8.The method of claim 1, wherein, in the formula, G represents the correlation degree, max (Xa, Xb) represents the maximum value of Xa and Xb; Xa represents the number of times a commodity variety a appears in historical orders, Xb represents the number of times a commodity variety b appears in the historical orders, and m represents the number of historical orders in which the commodity variety a and the commodity variety b appear together. The method further comprises the following steps: converting the commodity variety into a commodity vector representation; calculating the similarity between each two commodity vector representations by using a cosine similarity algorithm; For each of the commodity varieties, record the top preset number of similar commodity varieties with the highest similarity to the commodity variety, to obtain a similarity table; The matching and recommending of the target commodity in the target commodity variety according to the user label and the real-time recommendation label further includes: Obtaining the similar commodity variety corresponding to the target commodity from the similarity table; The matching and recommending of the target commodity in the target commodity variety and the similar commodity variety according to the user label and the recommendation label. 9.The method of claim 1, wherein, Further includes: Obtaining a preset large language model template, and filling the user label, the real-time recommendation label and the scene label into the large language model template to obtain a large language model prompt word; Inputting the large language model prompt word into a large language model to obtain a first target commodity; According to the user label and the real-time recommendation label, matching a second target commodity in the target commodity variety according to a preset rule; Calculating a first adaptation value of the first target commodity and a second adaptation value of the second target commodity; According to the comparison result of the first adaptation value and the second adaptation value, recommending the first target commodity or the second target commodity.

10. An electronic device, comprising: A computer program stored in the memory and running on the processor, when the processor executes the computer program, realizes each step of the above-mentioned order intelligent recommendation method.

Images