A method and device for constructing training samples of a recall model and a computer device

Abstract

This application provides a method, apparatus, and computer device for constructing training samples for a recall model. The method is applicable to shopping platforms and includes: obtaining at least one second product displayed after a user triggers a preset behavior for a first product on a preset page of the shopping platform; the first product and any second product have a similarity relationship; obtaining at least one third product in the shopping platform that matches the attribute information of the first product; constructing any product pair based on at least one second product and at least one third product, inputting the product pair and a prompt word into a large language model to obtain the similarity corresponding to the product pair; the prompt word is used to instruct the large language model to determine the similarity corresponding to the product pair from the product name and the ingredient list of the product pair; removing product pairs with similarity less than a second threshold to obtain similar product samples, the similar product samples are used to train the model to obtain a first recall model, and the first recall model is used to recall products with substitution relationships.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus and computer equipment for constructing training samples for a recall model. Background Technology

[0002] The Item-to-Item (I2I) recall model is the core model in the recall phase of a recommendation system. It calculates the similarity between items and recalls similar candidates based on the user's historical interactions. In other words, the I2I recall model provides a set of other items associated with the current item. One application scenario of the I2I recall model is recalling similar items. Similar items refer to items that are substitutable; for example, if the current item is a local cucumber, then both organic cucumbers and fresh cucumbers can be considered similar items.

[0003] Currently, when constructing samples for training recall models to retrieve similar products, training samples are usually constructed based on products under the same category or products of the same brand. However, this construction scheme cannot construct training samples like <fresh crayfish (fresh food category)> and <boxing shrimp (cooked food category)>, because the two products are completely different categories. It may also mistakenly use <millet and millet peppers> as training samples, because the similarity of product pairs is currently mainly determined based on product names, and the accuracy of product pair similarity needs to be improved.

[0004] Constructing more high-quality training samples is of great significance for improving the training effect of recall models. Summary of the Invention

[0005] This application provides a method, apparatus, and computer device for constructing training samples for a recall model, in order to improve the quantity and quality of training samples for the recall model.

[0006] In a first aspect, this application provides a method for constructing training samples for a recall model, the method being applicable to a shopping platform, the method comprising: Obtain at least one second product displayed after a user triggers a preset behavior on a preset page of the shopping platform for a first product; the first product and any second product are similar in nature. Obtain at least one third product from the shopping platform that matches the attribute information of the first product; Based on any product pair constructed from the at least one second product and the at least one third product, the product pair and the prompt word are input into a large language model to obtain the similarity corresponding to the product pair; the prompt word is used to instruct the large language model to determine the similarity corresponding to the product pair from the product name and the ingredient list of the product pair; Product pairs with similarity less than a second threshold are removed to obtain similar product samples. These similar product samples are used to train a model to obtain a first recall model, which is used to recall products with substitution relationships.

[0007] The products matching the attribute information of the first product on the shopping platform include products of the same category or brand as the first product. In addition, this application also obtains at least one second product displayed after a user triggers a preset behavior for the first product on a preset page of the shopping platform. Similar product samples constructed based on at least one second product and at least one third product increase the number of similar product samples compared to the existing solutions described in the background section. Furthermore, after constructing product pairs based on at least one second product and at least one third product, this application also determines the similarity of product pairs from product names and ingredient lists using a large language model, removing products with low similarity from the product pairs, thus avoiding the situation where "<millet, millet pepper>" is used as a similar product sample and improving the quality of similar product samples.

[0008] In one possible design, acquiring at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product includes: Obtain at least one searched product displayed after a user triggers a search for a first product on the search page of the shopping platform; and designate the searched product whose relevance to the first product is greater than a first threshold and which is of interest to the user as the second product; and / or The similar products displayed on the similar products page and which are followed by the user are obtained as the second products. The similar products page is the page that the user enters after clicking the option to view similar products for the first product on the shopping platform. Being noticed by the user includes at least one action such as clicking, adding to cart, or making a purchase.

[0009] For at least one searched product, products whose relevance to the first product is greater than a first threshold and which have been clicked, added to the shopping cart, or purchased by the user are designated as second products; for products displayed on similar product pages, products that have been clicked, added to the shopping cart, or purchased by the user are designated as second products; in other words, regardless of whether it is a searched product or a product displayed on a similar product page, this application determines the second product based on user behavior feedback data; thus, the similarity between at least one second product is improved, thereby improving the quality of subsequent similar product samples constructed based on at least one second product and at least one third product.

[0010] In one possible design, the cue word is used to instruct a large language model to determine the similarity of the product pair from the product names and ingredient lists of the two products included in the product pair, including: The prompt words are used to instruct the large language model to determine the first similarity of the product pairs from the core words included in the product names; The prompt words are used to instruct the large language model to determine the second similarity of the product pair from the main ingredients in the product ingredient list; The prompt words are used to instruct the large language model to determine the third similarity of the product pair from the auxiliary ingredients in the product ingredient list; Based on the first similarity, the second similarity, and the third similarity, the similarity of the corresponding product pairs is determined.

[0011] In addition to determining the first similarity of product pairs based on product names, this application also determines the second and third similarities of product pairs based on the ingredient lists of the products. Finally, it determines the corresponding similarity of product pairs based on the first to third similarities, effectively eliminating samples such as "millet, millet pepper" and improving the quality of the final obtained similar product samples.

[0012] In one possible design, the method further includes: Obtain at least one fourth product displayed after the user clicks on the shopping platform to view the bundled products option for the first product; Obtain at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform; A sample of paired products is determined by constructing at least one fourth product and at least one fifth product. The sample of paired products is used to train a model to obtain a second recall model. The second recall model is used to recall products that have a paired relationship.

[0013] In one possible design, before retrieving at least one fifth product that appears simultaneously with the first product from the shopping platform's historical orders, the method further includes: For any order containing the first product for any user, at least one product is removed based on the frequency of occurrence of each product in the order, and the remaining product is designated as the fifth product. The frequency of a product's appearance is determined by the number of times the product appears in the orders of each user on the shopping platform. The probability of a product being removed is positively correlated with its frequency of appearance.

[0014] The at least one item is removed based on its frequency of occurrence in the order, including: For any one of the at least one products, a rejection threshold is set for that product; the rejection threshold is a random number. If the frequency of occurrence of the product is greater than the removal threshold, the product will be removed from the at least one product.

[0015] Taking a grocery shopping platform as an example, versatile complementary items such as scallions, ginger, and garlic may appear in most orders, which greatly interferes with the construction of the complementary item sample. This application removes items from the order based on the frequency of their appearance, which can effectively remove versatile complementary items from the order, thereby improving the quality of the complementary item sample and the training effect of the second recall model.

[0016] Secondly, this application also provides a device for constructing training samples for recall models, the device being deployed on a shopping platform, the device comprising: a processing unit and a transceiver unit; The transceiver unit is used to acquire at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product; the first product and any second product are similar in nature; The transceiver unit is also used to acquire at least one third product in the shopping platform that matches the attribute information of the first product; The processing unit is configured to input any product pair constructed based on the at least one second product and the at least one third product into a large language model to obtain the similarity of the product pair; the prompt word is used to instruct the large language model to determine the similarity of the product pair from the product name and the ingredient list of the product pair. The processing unit is further configured to remove product pairs with similarity less than a second threshold to obtain similar product samples. The similar product samples are used to train a model to obtain a first recall model, which is used to recall products with substitution relationships.

[0017] In one possible design, the transceiver unit is used to acquire information about at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product. Specifically, this is used for: Obtain at least one searched product displayed after a user triggers a search for a first product on the search page of the shopping platform; and designate the searched product whose relevance to the first product is greater than a first threshold and which is of interest to the user as the second product; and / or The similar products displayed on the similar products page and which are followed by the user are obtained as the second products. The similar products page is the page that the user enters after clicking the option to view similar products for the first product on the shopping platform. Being noticed by the user includes at least one action such as clicking, adding to cart, or making a purchase.

[0018] In one possible design, the cue word is used to instruct a large language model to determine the similarity of the product pair from the product names and ingredient lists of the two products included in the product pair, including: The prompt words are used to instruct the large language model to determine the first similarity of the product pairs from the core words included in the product names; The prompt words are used to instruct the large language model to determine the second similarity of the product pair from the main ingredients in the product ingredient list; The prompt words are used to instruct the large language model to determine the third similarity of the product pair from the auxiliary ingredients in the product ingredient list; Based on the first similarity, the second similarity, and the third similarity, the similarity of the corresponding product pairs is determined.

[0019] In one possible design, the transceiver unit is also configured to acquire at least one fourth product displayed on the shopping platform after the user clicks on the option to view complementary products for the first product. The transceiver unit is also used to obtain at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform; The processing unit is further configured to determine a paired product sample constructed from the at least one fourth product and the at least one fifth product, the paired product sample being used to train a model to obtain a second recall model, the second recall model being used to recall products with a paired relationship.

[0020] In one possible design, before retrieving at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform, the processing unit is further configured to: For any order containing the first product for any user, at least one product is removed based on the frequency of occurrence of each product in the order, and the remaining product is designated as the fifth product. The frequency of a product's appearance is determined by the number of times the product appears in the orders of each user on the shopping platform. The probability of a product being removed is positively correlated with its frequency of appearance.

[0021] In one possible design, the processing unit, when removing at least one item from the order based on the frequency of occurrence of each item, specifically performs the following: For any one of the at least one products, a rejection threshold is set for that product; the rejection threshold is a random number. If the frequency of occurrence of the product is greater than the removal threshold, the product will be removed from the at least one product.

[0022] Thirdly, this application also provides a computer device, the device comprising: a processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method described in the first aspect above.

[0023] Fourthly, this application also provides a computer-readable storage medium comprising a program that, when executed on a device, causes the device to perform the method as described in any one of the first aspects above.

[0024] Fifthly, this application also provides a computer program product, the computer program product comprising a computer program that, when executed by a processor, implements the method described in the first aspect above. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic diagram illustrating the method for constructing training samples for the recall model provided in this application embodiment; Figure 2 This is a schematic diagram illustrating the process of obtaining similar product samples provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the process of obtaining product samples in an embodiment of this application. Figure 4 Schematic diagram of the structure of the device for constructing training samples for the recall model provided in the embodiments of this application. Figure 1 ; Figure 5 Schematic diagram of the structure of the device for constructing training samples for the recall model provided in the embodiments of this application. Figure 2 ; Figure 6 This is a schematic diagram of the computer device structure provided in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] The application scenarios described in this application are for the purpose of more clearly illustrating the technical solutions of this application, and do not constitute a limitation on the technical solutions provided in this application. Those skilled in the art will understand that with the emergence of new application scenarios, the technical solutions provided in this application are also applicable to similar technical problems. In the description of this application, unless otherwise stated, "multiple" means two or more.

[0029] In addition to recalling similar products as described in the background section, another common application scenario for recall models is recalling complementary products of the current product. Complementary products refer to products that have a certain pairing relationship; for example, assuming the current product is tomatoes, eggs and beef can both be considered complementary products of the current product.

[0030] Currently, when constructing samples for training recall models that retrieve paired products, training samples are typically built based on click behavior within a user session. However, training samples constructed in this way are prone to noise interference. For example, suppose a user clicks on four products in the same session: tomato, egg, cucumber, and potato. Existing solutions might construct training samples based on these four products as follows: <tomato, egg>, <tomato, cucumber>, <tomato, potato>, <egg, cucumber>, <egg, potato>, and <cucumber, potato>. While <tomato, egg> and <egg, cucumber> do have a pairing relationship, <tomato, cucumber>, <tomato, potato>, <egg, potato>, and <cucumber, potato> do not, constituting noise interference and affecting the training effect of the recall model.

[0031] To address the problems existing in the construction schemes of similar product samples and combined product samples, this application proposes the following... Figure 1 The method shown illustrates the construction of training samples for the recall model, which can be applied to shopping platforms. For example... Figure 1 As shown, the executing entity of this method can be a server, a chip within the server, or a functional module within the server; this application does not limit this. The following explanation uses a server as the executing entity. The method includes: Step 101: Obtain at least one second product displayed after the user triggers a preset behavior for the first product on a preset page of the shopping platform; the first product and any second product have a similar relationship.

[0032] For example, shopping platforms include, but are not limited to, Taobao, JD.com, and Dingdong Maicai, and this application does not limit the shopping platform. One source of at least one second product is to obtain similar products displayed on a similar products page that are being viewed by the user. The similar products page is the page the user enters after clicking the "view similar products" option on the shopping platform for the first product. This similar products page can be the page accessed after clicking the "view similar products" option provided by the shopping platform while browsing products, or it can be the page accessed after clicking the "view similar products" option provided by the shopping platform for expired products in the shopping cart. Another source of at least one second product is to obtain at least one search product displayed after the user triggers a search for the first product on the shopping platform's search page. Search products with a relevance greater than a first threshold and being viewed by the user are used as second products. Being viewed by the user includes at least one instance of clicking, adding to the shopping cart, or being in the process of purchasing.

[0033] For example, assuming the first product is an Apple iPhone, after a user enters "Apple iPhone" into the search page of a shopping platform and triggers a search, the page will display at least one search result obtained from the search for "Apple iPhone." This at least one search result may include various models of Apple iPhones, Huawei phones, OPPO phones, and VIVO phones, and may also include some Apple phone cases. Search results for products whose relevance to the first product is greater than a first threshold and which have been clicked, added to cart, or purchased by the user are designated as the second product. The relevance between the search result and the first product can be obtained by calculating the similarity between the two products, or other methods of determining relevance can be used; this application does not limit this approach. If the user clicks on Apple phone case A among the at least one search result, since the relevance between Apple phone case A and the first product "Apple iPhone" is lower than the first threshold, Apple phone case A will not be designated as the second product. If a user clicks on the Apple iPhone 17 Pro in at least one of the searched products, the Apple iPhone 17 Pro can be considered as the second product because its relevance to the first product "Apple iPhone" is greater than the first threshold.

[0034] Thus, any one of the at least one second product obtained is similar to the first product. This similarity can be understood as any second product being a substitute for the first product. For example, whole milk and skim milk are similar, as are peanuts and pork head meat, because both are considered appetizers.

[0035] Step 102: Obtain at least one third product from the shopping platform that matches the attribute information of the first product.

[0036] After obtaining at least one second product, at least one third product from the shopping platform that matches the attribute information of the first product can also be obtained. The attribute information of the first product can be its category, such as fresh produce, dairy products, or fruits; it can also be the brand of the first product; or it can be the description of the first product, such as "the first product is rich in vitamin C" or "the first product is used to treat symptoms such as fever and runny nose." This application does not limit the attribute information of the first product. Obtaining at least one third product from the shopping platform that matches the attribute information of the first product can further increase the number of subsequent similar product samples.

[0037] Step 103: Based on any product pair constructed from at least one second product and at least one third product, input the product pair and the prompt word into the large language model to obtain the similarity of the product pair; the prompt word is used to instruct the large language model to determine the similarity of the product pair from the product name and the ingredient list of the product pair.

[0038] After obtaining at least one second product and at least one third product, multiple product pairs can be constructed based on the second and third products. In order to avoid the existence of product pairs such as "<millet, millet pepper>" which do not belong to similar product samples in the final similar product samples, this application can input any one of the constructed product pairs and the prompt word into the large language model to obtain the similarity of the product pairs.

[0039] Among them, the cue words are used to instruct the large language model to determine the similarity of corresponding product pairs from the product names and ingredient lists, specifically: (1) The prompt words are used to instruct the large language model to determine the first similarity of the product pair from the core words included in the product name; for example, the value of the first similarity can be in the range of 0 to 3. If the core words of the product names of the two products included in the product pair are the same, the first similarity of the product pair can be determined as 3; if there are no identical words or synonyms between the product names, the first similarity of the product pair can be determined as 0. Taking the product pair <local cucumber, organic cucumber> as an example, the core words of the product names of the two products included in this product pair are the same, both being cucumber, so the first similarity of this product pair can be determined as 3.

[0040] (2) The prompt words are used to instruct the large language model to determine the second similarity of the product pair from the main ingredients in the product ingredient list. For example, the value range of the second similarity can be 0 to 2. If the main ingredients in the product ingredient lists of the two products included in the product pair are the same, the second similarity of the product pair can be determined as 2. If the main ingredients in the product ingredient lists of the two products included in the product pair are completely different, the second similarity of the product pair can be determined as 0. Taking the product pair <cucumber, pickled cucumber> as an example, the main ingredients in the product ingredient lists of the two products included in the product pair are the same, so the second similarity of the product pair can be determined as 2.

[0041] (3) The prompt words are also used to instruct the large language model to determine the third similarity of the product pair from the auxiliary ingredients in the product ingredient list. For example, the value of the third similarity can be 0 to 1. If the auxiliary ingredients in the ingredient lists of the two products included in the product pair are the same, the third similarity of the product pair can be determined as 1. If the auxiliary ingredients in the ingredient lists of the two products included in the product pair are completely different, the third similarity of the product pair can be determined as 0. For example, taking the product pair <wasabi-flavored rice crust, wasabi-flavored potato chips> as an example, the auxiliary ingredients in the ingredient lists of the two products included in this product pair are the same, both being wasabi-flavored seasoning, so the third similarity of the product pair can be determined as 1.

[0042] It should be noted that the range of values ​​for the first to third similarity mentioned above is only an example. The range of values ​​for the first to third similarity can be the same or different, and this application does not limit this.

[0043] After determining the first similarity, second similarity, and third similarity, the similarity of the corresponding product pair is determined based on the first similarity, second similarity, and third similarity. Specifically, the maximum value among the first and third similarities can be used as the similarity of the corresponding product pair, or the first and third similarities can be weighted and summed, and the weighted sum can be used as the similarity of the corresponding product pair. This application does not limit the method of determining the similarity of the corresponding product pair based on the first and third similarities.

[0044] For example, suppose that among multiple product pairs constructed based on at least one second product and at least one third product, there are <tomato-flavored potato chips, barbecue-flavored potato chips>. The two products in this product pair not only have the same core words in their product names, but also the same main ingredients in their ingredient lists. However, the auxiliary ingredients in their ingredient lists are not exactly the same. The large language model determines the first similarity of this product pair to be 3, the second similarity to be 2, and the third similarity to be 0.3. If the maximum value among the first to third similarities is taken as the similarity of the product pair, then the similarity of <tomato-flavored potato chips, barbecue-flavored potato chips> is 3.

[0045] In addition to determining the first similarity of product pairs based on product names, this application also determines the second and third similarities of product pairs based on product ingredient lists. Finally, it determines the corresponding similarity of product pairs based on the first to third similarities, which improves the accuracy of similarity determination for product pairs such as "<millet, millet pepper>" and enables more accurate removal of product pairs with similarity less than the second threshold.

[0046] Step 104: Remove product pairs with similarity less than the second threshold to obtain similar product samples. The similar product samples are used to train the model to obtain the first recall model. The first recall model is used to recall products with substitution relationships.

[0047] Step 103 determines the similarity of each product pair among multiple product pairs constructed based on at least one second product and at least one third product. Product pairs with similarity less than the second threshold are removed, and only product pairs with high similarity are retained. This improves the quality of similar product samples and thus enhances the training effect of the first recall model.

[0048] Steps 101 to 104 above describe the construction of similar product samples. The sources of similar product samples can also be found in [reference needed]. Figure 2 This application also provides, for example, Figure 3 The method for constructing the product combination sample shown is as follows: At least one fourth product is displayed after a user clicks on the "View Product Combinations" option for the first product on the shopping platform; at least one fifth product appears simultaneously with the first product in the historical orders of the shopping platform, i.e., at least one fifth product appears in the same order as the first product in the historical orders of the shopping platform; finally, a product combination sample is constructed by determining at least one fourth product and at least one fifth product. This product combination sample is used to train the model to obtain the second recall model, which is used to recall products with a product combination relationship.

[0049] Before obtaining at least one fifth product that appears simultaneously with the first product in historical orders from the shopping platform, this application removes products from historical orders to reduce noise interference in the paired product samples. This removal process borrows from an idea similar to Inverse Document Frequency (IDF). Specifically, for any order containing the first product for any user, at least one product is removed based on the frequency of occurrence of each of the at least one product included in the order, and the remaining product is designated as the fifth product. The frequency of occurrence of a product is determined by the number of times the product appears in the orders of each user on the shopping platform, and the probability of a product being removed is positively correlated with its frequency of occurrence; that is, the higher the frequency of occurrence of a product, the higher the probability of it being removed. When removing at least one product based on its frequency of occurrence, a removal threshold is set for any one of the at least one products. This threshold is a random number. If the frequency of occurrence of the product is greater than the removal threshold, the product is removed from the at least one product.

[0050] For example, suppose the first product is tomato. Taking user A on a shopping platform as an example, user A has 20 orders including tomatoes. One order includes the following items: tomatoes, garlic, tofu, eggs, milk, toast, and scallions. Assume the frequency of tomatoes is 70%, garlic 78%, tofu 63%, eggs 78%, milk 80%, toast 60%, and scallions 82%. For tomatoes, a random number ① is determined as the elimination threshold. Assuming random number ① is 0.8, since tomatoes... The current frequency is 70%, which is less than random number ①. Therefore, tomatoes are retained in this order. Similarly, random number ② is determined as the removal threshold for garlic. Assuming random number ② is 0.7, since the frequency of garlic is 78%, which is greater than random number ②, garlic is removed. ... Random number ⑦ is determined as the removal threshold for scallions. Assuming random number ⑦ is also 0.8, since the frequency of scallions is 82%, which is greater than 0.8, scallions are removed. In the end, the items retained in the fifth item of user A's order are tomatoes, tofu, eggs, milk, and toast. Versatile items like garlic and scallions have been removed.

[0051] For the processing method of User A's remaining 19 orders, please refer to the processing method for that order. For the processing method of other users' orders on the shopping platform, please refer to the processing method for User A's orders; it will not be repeated here. After all orders of all users on the shopping platform are processed, at least one fifth item will be obtained.

[0052] In addition, this application can also obtain some common matching products through a large language model, thereby increasing the number of matching product samples; it can also remove some products that cannot be matched from the constructed product pairs through a large language model, such as <hairy crab, persimmon>, thereby improving the quality of matching product samples.

[0053] Before acquiring the fifth product, this application removes products from the order based on their frequency of appearance, effectively eliminating generic matching products from the order. Subsequently, when constructing matching product samples based on at least one fourth product and at least one fifth product, the quality of the matching product samples is improved, enhancing the training effect of the second recall model, achieving accurate user recommendations, providing users with a better experience, and thus improving the conversion efficiency of the shopping platform.

[0054] After obtaining similar product samples and paired product samples, this application does not limit the selection of the training model. For example, similar product samples or paired product samples can be input into the Swing model for training according to the format requirements of the Swing model, thereby obtaining a first recall model or a second recall model.

[0055] Figure 4 and Figure 5 This is a schematic diagram of a possible apparatus for constructing recall model training samples according to embodiments of this application. These apparatuses for constructing recall model training samples can be used to implement the server functionality described in the above method embodiments, and thus also achieve the beneficial effects of the above method embodiments.

[0056] like Figure 4 As shown, the apparatus 400 for constructing training samples for the recall model is deployed on a shopping platform and includes a processing unit 410 and a transceiver unit 420. The apparatus 400 for constructing training samples for the recall model is used to implement the above-mentioned... Figure 1 The server functionality is illustrated in the method embodiment shown.

[0057] When the device 400 for constructing training samples for the recall model is used to implement Figure 1 The server function in the method embodiment shown is as follows: The transceiver unit 420 is used to acquire at least one second product displayed after a user triggers a preset behavior on a preset page of the shopping platform for a first product; the first product and any second product have a similar relationship; The transceiver unit 420 is also used to acquire at least one third product in the shopping platform that matches the attribute information of the first product; The processing unit 410 is configured to input any product pair constructed based on the at least one second product and the at least one third product into a large language model to obtain the similarity corresponding to the product pair; the prompt word is used to instruct the large language model to determine the similarity corresponding to the product pair from the product name and the ingredient list of the product pair; The processing unit 410 is further configured to remove product pairs with similarity less than a second threshold to obtain similar product samples. The similar product samples are used to train a model to obtain a first recall model. The first recall model is used to recall products with substitution relationships.

[0058] In one possible design, the transceiver unit 420 is used to acquire information about at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product. Specifically, it is used for: Obtain at least one searched product displayed after a user triggers a search for a first product on the search page of the shopping platform; and designate the searched product whose relevance to the first product is greater than a first threshold and which is of interest to the user as the second product; and / or The similar products displayed on the similar products page and which are followed by the user are obtained as the second products. The similar products page is the page that the user enters after clicking the option to view similar products for the first product on the shopping platform. Being noticed by the user includes at least one action such as clicking, adding to cart, or making a purchase.

[0059] In one possible design, the cue word is used to instruct a large language model to determine the similarity of the product pair from the product names and ingredient lists of the two products included in the product pair, including: The prompt words are used to instruct the large language model to determine the first similarity of the product pairs from the core words included in the product names; The prompt words are used to instruct the large language model to determine the second similarity of the product pair from the main ingredients in the product ingredient list; The prompt words are used to instruct the large language model to determine the third similarity of the product pair from the auxiliary ingredients in the product ingredient list; Based on the first similarity, the second similarity, and the third similarity, the similarity of the corresponding product pairs is determined.

[0060] In one possible design, the transceiver unit 420 is also configured to acquire at least one fourth product displayed on the shopping platform after the user clicks on the option to view complementary products for the first product. The transceiver unit 420 is also used to obtain at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform; The processing unit 410 is further configured to determine a sample of paired products constructed from the at least one fourth product and the at least one fifth product, the sample of paired products being used to train a model to obtain a second recall model, the second recall model being used to recall products with a paired relationship.

[0061] In one possible design, before retrieving at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform, the processing unit 410 is further configured to: For any order containing the first product for any user, at least one product is removed based on the frequency of occurrence of each product in the order, and the remaining product is designated as the fifth product. The frequency of a product's appearance is determined by the number of times the product appears in the orders of each user on the shopping platform. The probability of a product being removed is positively correlated with its frequency of appearance.

[0062] In one possible design, the processing unit 410, when removing at least one item from the order based on the frequency of occurrence of each item, specifically performs the following: For any one of the at least one products, a rejection threshold is set for that product; the rejection threshold is a random number. If the frequency of occurrence of the product is greater than the removal threshold, the product will be removed from the at least one product.

[0063] For a more detailed description of the processing unit 410 and the transceiver unit 420, please refer to [link / reference needed]. Figure 1 The relevant descriptions in the method embodiments shown are directly obtained and will not be repeated here.

[0064] like Figure 5 As shown, the apparatus 500 for constructing training samples for the recall model includes a processor 510 and an interface circuit 520. The processor 510 and the interface circuit 520 are coupled to each other. It is understood that the interface circuit 520 can be a transceiver or an input / output interface. Optionally, the apparatus 500 for constructing training samples for the recall model may also include a memory 530 for storing instructions executed by the processor 510, or storing input data required by the processor 510 to execute instructions, or storing data generated after the processor 510 executes instructions.

[0065] When the device 500 for constructing training samples for the recall model is used to implement Figure 1 In the method shown, the processor 510 is used to implement the functions of the processing unit 410, and the interface circuit 520 is used to implement the functions of the transceiver unit 420.

[0066] Based on the same technical concept, embodiments of this application provide a computer device, such as... Figure 6 As shown, it includes at least one processor chip 601 and a memory 602 connected to at least one processor chip. In this embodiment, the specific connection medium between the processor chip 601 and the memory 602 is not limited. Figure 6 Taking the connection between the processor chip 601 and the memory 602 via a bus as an example, the bus can be divided into address bus, data bus, control bus, etc.

[0067] In this embodiment, the memory 602 stores instructions that can be executed by at least one processor chip 601. By executing the instructions stored in the memory 602, the at least one processor chip 601 can perform the steps of the above-described method for constructing training samples for recalling models.

[0068] The processor chip 601 serves as the control center of the computer device. It connects to various parts of the device via various interfaces and lines, controlling the chip by running or executing instructions stored in the memory 602 and accessing data stored in the memory 602. Optionally, the processor chip 601 may include one or more processing units. The processor chip 601 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may not be integrated into the processor chip 601. In some embodiments, the processor chip 601 and the memory 602 may be implemented on the same chip; in other embodiments, they may be implemented on separate chips.

[0069] The processor chip 601 can be a general-purpose processor, such as a graphics processing unit (GPU), general-purpose computing on graphics processing units (GPGPU), central processing unit (CPU), digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0070] Memory 602, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 602 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory 602 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer device, but is not limited thereto. In the embodiments of this application, memory 602 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0071] Based on the same technical concept, embodiments of this application provide a computer-readable storage medium storing a computer program executable by a computer device. When the program is run on the computer device, it causes the computer device to perform the steps of the above-described method for constructing training samples for recall models.

[0072] Based on the same technical concept, this application provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions, which, when executed by a computer device, cause the computer device to perform the steps of the above-described method for constructing training samples for recall models.

[0073] Those skilled in the art will understand that embodiments of this application can be provided as methods or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer apparatus or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer device or other programmable data processing equipment to cause a series of operational steps to be performed on the computer device or other programmable equipment to produce a process implemented by the computer device, thereby providing instructions that execute on the computer device or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0077] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for constructing a training sample of a recall model, characterized in that, The method is applicable to shopping platforms, and the method includes: Obtain at least one second product displayed after a user triggers a preset behavior on a preset page of the shopping platform for a first product; the first product and any second product are similar in nature. Obtain at least one third product from the shopping platform that matches the attribute information of the first product; Based on any product pair constructed from the at least one second product and the at least one third product, the product pair and the prompt word are input into a large language model to obtain the similarity corresponding to the product pair; the prompt word is used to instruct the large language model to determine the similarity corresponding to the product pair from the product name and the ingredient list of the product pair; Product pairs with similarity less than a second threshold are removed to obtain similar product samples. These similar product samples are used to train a model to obtain a first recall model, which is used to recall products with substitution relationships.

2. The method of claim 1, wherein, Obtain at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product, including: Obtain at least one searched product displayed after a user triggers a search for a first product on the search page of the shopping platform; and designate the searched product whose relevance to the first product is greater than a first threshold and which is of interest to the user as the second product; and / or The similar products displayed on the similar products page and which are followed by the user are obtained as the second products. The similar products page is the page that the user enters after clicking the option to view similar products for the first product on the shopping platform. Being noticed by the user includes at least one action such as clicking, adding to cart, or making a purchase.

3. The method of claim 1, wherein, The cue words are used to instruct the large language model to determine the similarity of the product pair from the product names and ingredient lists of the two products included in the product pair, including: The prompt words are used to instruct the large language model to determine the first similarity of the product pairs from the core words included in the product names; The prompt words are used to instruct the large language model to determine the second similarity of the product pair from the main ingredients in the product ingredient list; The prompt words are used to instruct the large language model to determine the third similarity of the product pair from the auxiliary ingredients in the product ingredient list; Based on the first similarity, the second similarity, and the third similarity, the similarity of the corresponding product pairs is determined.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain at least one fourth product displayed after the user clicks on the shopping platform to view the bundled products option for the first product; Obtain at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform; A sample of paired products is determined by constructing at least one fourth product and at least one fifth product. The sample of paired products is used to train a model to obtain a second recall model. The second recall model is used to recall products that have a paired relationship.

5. The method of claim 4, wherein, Before obtaining at least one fifth product that appears simultaneously with the first product in the historical orders of the shopping platform, the method further includes: For any order containing the first product for any user, at least one product is removed based on the frequency of occurrence of each product in the order, and the remaining product is designated as the fifth product. The frequency of a product's appearance is determined by the number of times the product appears in the orders of each user on the shopping platform. The probability of a product being removed is positively correlated with its frequency of appearance.

6. The method of claim 5, wherein, The at least one item is removed based on its frequency of occurrence in the order, including: For any one of the at least one products, a rejection threshold is set for that product; the rejection threshold is a random number. If the frequency of occurrence of the product is greater than the removal threshold, the product will be removed from the at least one product.

7. A device for constructing a training sample of a recall model, characterized by comprising: The device is deployed on a shopping platform; the device includes: a processing unit and a transceiver unit; The transceiver unit is used to acquire at least one second product displayed after a user triggers a preset action on a preset page of the shopping platform for a first product; the first product and any second product are similar in nature; The transceiver unit is also used to acquire at least one third product in the shopping platform that matches the attribute information of the first product; The processing unit is configured to input any product pair constructed based on the at least one second product and the at least one third product into a large language model to obtain the similarity of the product pair; the prompt word is used to instruct the large language model to determine the similarity of the product pair from the product name and the ingredient list of the product pair. The processing unit is further configured to remove product pairs with similarity less than a second threshold to obtain similar product samples. The similar product samples are used to train a model to obtain a first recall model, which is used to recall products with substitution relationships.

8. A computer device, comprising: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterised in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1-6.

Images