Method for service recommendation based on identified user intention, computing device and storage medium

Abstract

The embodiment of the invention relates to a method for service recommendation based on an identified user intention, computing equipment and a storage medium. The method comprises the following steps: respectively performing scalar retrieval and vector retrieval in a keyword database according to current input information of a user; sorting based on the scalar retrieval result and the vector retrieval result to obtain a preliminary intention tendency sequence of the user; at least sorting the current input information of the user and the initial intention tendency of the user, and inputting an intention recognition model to determine the current intention of the user and a service type corresponding to the current intention through the intention recognition model; the intention recognition model is constructed based on a neural network model; and based on the service type corresponding to the current intention of the user, calling a recommendation tool corresponding to the service type so as to perform service recommendation for the current intention of the user. Therefore, the stability and accuracy of user intention recognition can be effectively improved, and accurate and efficient recognition of the user intention is realized.

Description

Technical Field

[0001] The present invention relates to the field of user intent perception, and more particularly to a method, a computing device, and a storage medium for performing service recommendations based on recognized user intent. Background Art

[0002] Traditional methods for making business recommendations based on identified user intent include directly identifying the intent of user input through large models (Large Language Model (LLM), Large Foundation Model (LFM)) or using a preset rule engine to identify the intent of user input. These methods have the following shortcomings: due to the divergence of the large model itself, it is difficult to stably identify the user's true intent; and the rule engine does not support semantic generalization, making it difficult to accurately identify the intent of the user's ambiguous input.

[0003] In summary, the traditional method for recommending services based on identified user intent has the disadvantage of being unable to stably and accurately identify the user's true intent. Summary of the Invention

[0004] In response to the above problems, the present invention provides a method, computing device and storage medium for business recommendation based on the identified user intent, which can effectively improve the stability and accuracy of user intent identification and achieve accurate and efficient identification of user intent.

[0005] According to a first aspect of the present invention, a method for making business recommendations based on an identified user intention includes: performing scalar retrieval and vector retrieval in a keyword database for the user's current input information, respectively; sorting based on the scalar retrieval results and the vector retrieval results to obtain a preliminary intention tendency ranking of the user; at least sorting the user's current input information and the user's preliminary intention tendency, and inputting them into an intention recognition model to determine the user's current intention and the business type corresponding to the current intention via the intention recognition model; the intention recognition model is constructed based on a neural network model; and based on the business type corresponding to the user's current intention, calling a recommendation tool corresponding to the business type to make business recommendations for the user's current intention.

[0006] According to a second aspect of the present invention, a computing device is provided, comprising: at least one processing unit; at least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, enabling the computing device to perform the steps of the method according to the first aspect.

[0007] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a machine, the method according to the first aspect of the present invention is executed.

[0008] According to a fourth aspect of the present invention, there is further provided a computer program product, comprising a computer program, wherein when the computer program is executed by a machine, the method of the first aspect of the present invention is performed.

[0009] In some embodiments, the business type includes: product recommendation, application recommendation, activity recommendation, question inquiry and / or chat.

[0010] In some embodiments, the keyword database includes: a scalar database, which is constructed based on scalar data of product data, activity data, application data and / or problem data corresponding to the business type; and a vector database, which is constructed based on vectorized product data, activity data, application data and / or problem data corresponding to the business type.

[0011] In some embodiments, performing scalar retrieval and vector retrieval in a keyword database respectively includes: vectorizing the user's current input information to obtain the vectorized current input information; performing similarity retrieval in a vector database for the vectorized current input information; and identifying at least one business type and at least one preliminary intention tendency corresponding to the user's current input information based on the level of semantic similarity.

[0012] In some embodiments, sorting based on scalar retrieval results and vector retrieval results to obtain the user's preliminary intention tendencies includes: based on the scalar retrieval results, obtaining multiple preliminary intention tendencies of the user, so as to assign corresponding weights to the multiple preliminary intention tendencies; based on the vector retrieval results, obtaining multiple preliminary intention tendencies for users with semantic similarity higher than a predetermined similarity threshold, and assigning corresponding weights to them respectively; and merging the user's preliminary intention tendencies and corresponding weights obtained based on the scalar retrieval results and the vector retrieval results respectively, so as to sort the obtained user's preliminary intention tendencies from high to low based on the weights.

[0013] In some embodiments, the user's preliminary intention tendency and corresponding weight obtained by merging the scalar retrieval results and the vector retrieval results include: for the user's current input information, in response to confirming that the same preliminary intention tendency exists in both the scalar retrieval results and the vector retrieval results, excluding the vector retrieval results of the same preliminary intention tendency.

[0014] In some embodiments, at least the user's current input information and the user's preliminary intention tendency are sorted and input into the intention recognition model to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model, including: obtaining the user's current input information, the user's historical input information, the user's historical intention information and the user's preliminary intention tendency sorting, and inputting them into the intention recognition model; identifying the user's predicted intention tendency through the intention recognition model, the predicted intention tendency including: at least one intention tendency belonging to a predetermined intention tendency set, and / or at least one preliminary intention tendency; and sorting based on the probability of the user's predicted intention tendency to determine the user's current intention and the business type corresponding to the current intention.

[0015] Therefore, the present invention can combine the advantages of both scalars and vectors, construct a scalar database to perform scalar retrieval on the user's current input data, and construct a vector database to perform vector retrieval on the user's current input data; it integrates the high accuracy and high efficiency of scalar retrieval with the ability of vector retrieval to support generalization and similarity search; and then integrates the retrieval results of both scalar retrieval and vector retrieval, thereby providing more accurate input for the intent recognition model to guide it to more accurately and efficiently identify user intent and the business type corresponding to the intent.

[0016] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements.

[0018] Figure 1 A schematic diagram of a system for implementing a method for performing service recommendation based on identified user intention according to an embodiment of the present invention is shown.

[0019] Figure 2 A flow chart of a method for performing service recommendation based on identified user intent according to an embodiment of the present invention is shown.

[0020] Figure 3 A flow chart of a method for performing vector search in a keyword database according to an embodiment of the present invention is shown.

[0021] Figure 4 A flowchart of a method for obtaining a user's preliminary intention tendency ranking according to an embodiment of the present invention is shown.

[0022] Figure 5 A flowchart for determining a user's current intention and a service type corresponding to the current intention according to an embodiment of the present invention is shown.

[0023] Figure 6 A block diagram of an electronic device according to an embodiment of the present invention is shown.

[0024] Figure 7 A data flow framework diagram of an intent recognition method according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0025] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0026] As used herein, the term "including" and its variations represent open inclusion, i.e., "including but not limited to." Unless otherwise stated, the term "or" means "and / or." The term "based on" means "based at least in part on." The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment." The term "another embodiment" means "at least one additional embodiment." The terms "first," "second," etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0027] As described above, traditional methods for making business recommendations based on identified user intent include, for example, directly identifying the intent of user input information through large models (Large Language Model (LLM), Large Foundation Model (LFM)); for example, identifying the intent of user input information through a preset rule engine. The above methods have the following shortcomings: due to the divergence of the large model itself, it is difficult to stably identify the user's true intent; and the rule engine does not support semantic generalization, making it difficult to accurately identify the user's ambiguous input.

[0028] In summary, the traditional method for recommending services based on identified user intent has the disadvantage of being unable to stably and accurately identify the user's true intent.

[0029] In order to at least partially solve one or more of the above-mentioned problems and other potential problems, an exemplary embodiment of the present invention proposes a solution for making business recommendations based on the identified user intentions. In the solution of the present invention, first, a scalar search and a vector search are performed in a keyword database for the user's current input information; sorting is performed based on the scalar search results and the vector search results to obtain the user's preliminary intention tendency ranking; then, at least the user's current input information and the user's preliminary intention tendency are sorted and input into an intention recognition model to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model; wherein the intention recognition model is constructed based on a neural network model. The present invention also calls a recommendation tool corresponding to the business type based on the business type corresponding to the user's current intention, so as to make business recommendations based on the user's current intention.

[0030] Therefore, the present invention can identify user intentions by combining scalar retrieval and vector retrieval, integrating the high accuracy and high efficiency of scalar retrieval and the ability of vector retrieval to support semantic generalization, so that it can perform generalized similarity calculations for scenarios and keywords not covered by scalar retrieval with high accuracy and high efficiency; and integrate the user's preliminary intention tendencies obtained by scalar retrieval and vector retrieval, and input the user's preliminary intention tendencies into the intention recognition model, and obtain the user's preliminary intention tendencies through retrieval to constrain the thinking scope of the large model, preventing unstable results caused by excessive divergence, and thus can provide more accurate user intention recognition results based on scalar retrieval and vector retrieval combined with the intention recognition model; it avoids the instability of pure large model intention recognition, and does not require additional model training, reducing the cost of model customization; and can call the corresponding recommendation tool based on the identified user's current intention combined with the business scenario to make more accurate recommendations.

[0031] Figure 1 FIG. 1 is a schematic diagram of a system 100 (hereinafter referred to as system 100 ) for implementing a method for recommending services based on identified user intentions according to an embodiment of the present invention. Figure 1 As shown in FIG, the system 100 includes a computing device 110, a server 130, a network 140, and a user terminal (eg, Figure 1 The computing device 110, the server 130 and the user terminals may exchange data via a network 140 (eg, the Internet, a local area network or a wide area network).

[0032] Regarding user terminals (e.g. Figure 1The user terminals Y1, Y2, ..., Yn in the figure are, for example, mobile devices, personal terminals, desktop computers, smart watches, tablet computers, interactive devices, etc. Users can perform interactive operations on their display interfaces (such as browsing information, inputting information, searching for products, etc.). The user terminals exchange data with the server 130 through the network, and users input information through the user terminals.

[0033] Receive user input from server 130, identify user intent based on the input, and make recommendations based on user intent. Server 130 may have a keyword database deployed on it, or a keyword database deployed on a cloud storage device interacting with server 130. Server 130 may also have an intent recognition model deployed on it, or the intent recognition model may be deployed on computing device 110 interacting with server 130.

[0034] Regarding server 130, it can be, for example, an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud database, cloud storage, network services, cloud communications, middleware services, domain name services, security services, as well as big data and artificial intelligence platforms.

[0035] Regarding the server 130, a method for performing business recommendations based on the identified user intention according to an embodiment of the present invention is deployed thereon. For example, the server 130 can obtain input information about the user through the network 140 and send the user input information, or keywords extracted based on the user input information, to the computing device 110 for training the network, model, and system provided by the embodiment of the present invention for performing business recommendations based on the identified user intention.

[0036] Regarding the computing device 110, it is used, for example, to train the network model of the method for performing business recommendation based on the identified user intent of an embodiment of the present invention, to train and / or deploy an intent recognition model, to deploy a keyword database, etc. The computing device 110 may have one or more processing units, including dedicated processing units such as GPU, FPGA and ASIC, and general-purpose processing units such as CPU. The computing device 110 may be an integration of multiple physical servers, an integration of multiple processing units, etc. In addition, one or more virtual machines may also be running on each computing device 110. In some embodiments, the computing device 110 and the server 130 may be integrated together, or they may be set separately from each other. In some embodiments, the computing device 110, for example, includes an intent retrieval module 112, a preliminary intent acquisition module 114, an intent recognition module 116, and a business recommendation module 118.

[0037] Regarding the intention retrieval module 112 , it is used to perform scalar retrieval and vector retrieval in the keyword database for the user's current input information.

[0038] The preliminary intention acquisition module 114 is used to perform sorting based on the scalar retrieval results and the vector retrieval results to obtain the user's preliminary intention tendency ranking.

[0039] Regarding the intention recognition module 116, it is used to at least sort the user's current input information and the user's preliminary intention tendency, and input the intention recognition model to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model; the intention recognition model is constructed based on a neural network model.

[0040] The service recommendation module 118 is used to call a recommendation tool corresponding to a service type based on the service type corresponding to the user's current intention, so as to make a service recommendation based on the user's current intention.

[0041] Figure 2 1 is a flow chart of a method 200 for recommending services based on identified user intentions according to an embodiment of the present invention. Figure 1 The computing device 110 shown is executed, and may also be executed on Figure 6 The method 200 for recommending services based on the identified user intent may further include additional steps not shown and / or may omit steps shown, and the scope of the present invention is not limited in this respect.

[0042] In step 202 , the computing device 110 performs a scalar search and a vector search in the keyword database for the current input information of the user.

[0043] In some embodiments, the keyword database includes a scalar database and a vector database.

[0044] In some embodiments, the scalar database is constructed based on scalar data regarding product data, activity data, application data, and / or problem data corresponding to a business type.

[0045] In some embodiments, the business type includes: product recommendation, application recommendation, activity recommendation, question inquiry and / or chat.

[0046] Regarding the scalar database, a large number of keywords are stored in the scalar database. These keywords are directly stored in the form of character strings. For example, in this solution, based on the business type, several keywords under the same business type are stored in the form of scalars.

[0047] For example, if the business type is product recommendation, the stored keywords are "product ID (unique identifier), product name (such as milk tea, watermelon, mobile phone, tennis shoes), product category / subcategory (such as beverages, fruits, electronic devices, sports shoes), brand (such as Cha Yan Yue Se, Huawei, Xiaomi, Apple, etc.), model / specification (such as Sheng Sheng Oolong, Youlan Latte; 8424, Kirin; iPhone 15Pro Max 256GB, Xiaomi 14 White 512GB; Nike Vapor pro White Size 42)".

[0048] For example, if the business type is application recommendation, the stored keywords include "application ID (such as App Store ID, Google Play Package Name), application name (exact name, such as Genshin Impact, Honor of Kings, QQ, PDF Reader, Scallop, PS), application category / subcategory (such as games, social networking, tools, education, photography), developer / publisher (such as miHoYo, Tencent, Adobe), platform (such as iOS, Android, Windows, macOS), version number (current version), file size (xxGB), download / install volume (range or specific value), rating / rating (average, such as 4.5 stars), price (such as free, paid, in-app purchase), supported languages ​​(such as Japanese, Simplified Chinese), age rating (such as 4+, 12+, 17+), update time (specifically, the most recent update time), and keyword tags (core words describing the application's functions, such as weather forecast, food delivery, express delivery, shopping).

[0049] In some embodiments, the vector database is constructed based on vectorized product data, activity data, application data and / or problem data corresponding to the business type.

[0050] Regarding embedding, for example, embedding models (such as BERT, CLIP, Text2Vec) are used to convert text / image / structured data into floating-point vectors of fixed dimensions. For example, semantic models and / or image models are used to extract semantic and / or visual features from product data, activity data, application data, and / or question data corresponding to a business type. The extracted keywords related to the business type are then vectorized and stored in a vector database to facilitate vector retrieval based on the semantic similarity and image similarity of the product data, activity data, application data, and / or question data corresponding to the business type.

[0051] For example, the product data "Apple, Red Fuji, 10 yuan per catty" is vectorized, the activity information "Apple, apple, mobile phone" is vectorized, and the application data "Moji, weather forecast", "Meituan, takeout", "JD, shopping / takeout", and "PlayerUnknown's Battlegrounds, game, fps" are vectorized. It should be understood that data on various business types can be adjusted according to needs, and they will not be listed one by one here, but only a few examples are given.

[0052] For example, in different business scenarios, the same keyword contains one or more meanings. The meaning of the keyword in the user's intention is determined based on the keyword's associated keyword meaning and the user's input content. Regarding the same keyword containing one or more meanings, the meaning of the same keyword corresponding to the user's current input information is determined by the difference in business types, the meaning of other keywords associated with it, the user's current input information and / or the user's initial intention tendency. For example, does the "apple" entered by the user refer to "apple among fruits" or "Apple electronic devices."

[0053] It is worth noting that in different business scenarios, the same keyword may represent different meanings and correspond to different data content. For example, the keyword "apple" in the product recommendation business type and the keyword "apple" in the application recommendation business type may have different meanings and data content. The former may correspond to "apple fruit" or "Apple mobile phone", and the latter may correspond to "apple store".

[0054] In step 204 , the computing device 110 performs sorting based on the scalar search results and the vector search results to obtain a preliminary intention tendency ranking of the user.

[0055] For example, based on the scalar retrieval results and vector retrieval results, the three preliminary intention tendencies of the user are obtained as follows: Figure 1 , 70%; preliminary Figure 2 , 90%; preliminary Figure 3 , 60%;", according to the weights (or probabilities) of these preliminary intentions, sort them to get the user's preliminary intention tendency ranking "Preliminary Intention Figure 2 , 90%; preliminary Figure 1 , 70%; preliminary Figure 3 , 60%".

[0056] Therefore, the present invention can combine scalar data and vector data, perform scalar search on the user's current input data in order to obtain accurate search results; perform vector search at the same time to obtain search results with high similarity; combine the high accuracy and high efficiency of scalar search with the ability of vector search to support semantic generalization; and combine the search results of scalar search and vector search to obtain the user's preliminary intention tendency ranking, thereby providing more accurate input for the intention recognition model to guide it to more accurately and efficiently recognize user intentions.

[0057] The following will be combined Figure 3 Detailed description of the method for performing vector search in a keyword database; and Figure 4 The method for obtaining the user's preliminary intention tendency ranking is described in detail; it will not be repeated here.

[0058] In step 206, the computing device 110 at least sorts the user's current input information and the user's preliminary intention tendency, and inputs the information into the intention recognition model to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model; the intention recognition model is constructed based on a neural network model.

[0059] For example, if the user's current intention is to buy apples, the corresponding business type is product recommendation; if the user's current intention is to buy a mobile phone, the corresponding business type is product recommendation; if the user's current intention is to order takeout, the corresponding business type is application recommendation and / or product recommendation, such as recommending takeout applications, recommending takeout product links, etc.; if the user's current intention is to consult a question, the corresponding business type is problem consultation, etc.

[0060] The following will be combined Figure 5 The method for determining the user's current intention and the business type corresponding to the current intention is described in detail and will not be repeated here.

[0061] In step 208 , the computing device 110 calls a recommendation tool corresponding to the business type based on the business type corresponding to the current intention of the user, so as to make a business recommendation based on the current intention of the user.

[0062] For example, if the business type is product recommendation, relevant products will be recommended to users through the product search tool; if the business type is activity recommendation, relevant products will be recommended to users through the activity search tool; if the business type is application recommendation, relevant products will be recommended to users through the application search tool; if the business type is chatting, chatting with users will be conducted through the chat tool; if the business type is question consultation, relevant questions will be answered for users through the question-and-answer tool.

[0063] Therefore, by determining the business type corresponding to the intent, the user's intent can be determined more accurately, and based on the user's intent, the adapted business tools can be called to make more suitable business recommendations for the user.

[0064] In the above scheme, the present invention combines the advantages of both scalars and vectors, constructs a scalar database to perform scalar retrieval on the user's current input data, and constructs a vector database to perform vector retrieval on the user's current input data; it integrates the high accuracy and high efficiency of scalar retrieval with the generalization support and similarity search capabilities of vector retrieval; and then integrates the retrieval results of both scalar retrieval and vector retrieval, thereby providing more accurate input for the intent recognition model to guide it to more accurately and efficiently identify user intent and the business type corresponding to the intent.

[0065] In addition, the above solution also provides multiple business tools, corresponding to each business type, so that based on the identified user intention, the business tool corresponding to the business type can be called to make more suitable business recommendations for the user. Compared with the traditional business recommendation method that only uses one tool to make multiple business recommendations, the above solution can obtain more efficient and targeted business recommendations, so that the recommendation results can better match the user's current input content and be more in line with the user's true intention.

[0066] Figure 3 FIG. 3 is a flow chart showing a method 300 for performing vector search in a keyword database according to an embodiment of the present invention. Figure 1 The computing device 110 shown is executed, and may also be executed on Figure 6 The method 300 is executed at the electronic device 600. It should be understood that the method 300 for performing vector search in a keyword database may further include additional steps not shown and / or may omit the steps shown, and the scope of the present invention is not limited in this respect.

[0067] In step 302 , the computing device 110 vectorizes the user's current input information to obtain vectorized current input information.

[0068] Regarding the vectorization of the user's current input information, for example, using an embedding model (such as BERT, CLIP, Text2Vec) to convert the user's current input information data into vector information, so as to facilitate similarity retrieval in the vector database. The user's current input information can be one or more of text, pictures and voice. If the user enters voice information, the voice will be converted into text; it supports multiple types of user input information.

[0069] In step 304 , the computing device 110 performs a similarity search in the vector database for the vectorized current input information.

[0070] In step 306 , the computing device 110 identifies at least one business type and at least one preliminary intention tendency corresponding to the user's current input information based on the semantic similarity.

[0071] Regarding scalar searches within a keyword database, for example, after receiving the user's current input, scalar and vector searches are performed on the user's current input. Scalar searches do not require vectorization of the user's current input. During scalar searches, operations such as precise search and fuzzy matching are performed solely on the user's current input. For example, for the user's current input, "What's good takeout today? I want to order lunch?", the scalar search process involves performing full-text precise search, keyword precise search (e.g., for "takeout" and "lunch"), and fuzzy matching searches.

[0072] Therefore, the user's current input information will be subjected to scalar retrieval and vector retrieval respectively. When performing scalar retrieval, there is no need to vectorize the user's current input information, and it can be directly searched in the scalar database in its original form and / or keywords extracted based on the original form, thereby improving the retrieval efficiency; while for vector retrieval, the user's current input information needs to be vectorized so that similarity calculation can be performed, so as to retrieve results with high similarity to the user's current input information in the vector database to determine the user's initial intention, thereby solving the limitations of the scalar database and supplementing the results that cannot be retrieved by scalar retrieval.

[0073] Figure 4 The flowchart of the method 400 for obtaining the preliminary intention tendency ranking of the user according to an embodiment of the present invention is shown. The method 400 for obtaining the preliminary intention tendency ranking of the user can be as follows: Figure 1 The computing device 110 shown is executed, and may also be executed on Figure 6 It is executed at the electronic device 600 shown. It should be understood that the method 400 for obtaining the user's preliminary intention tendency ranking may further include additional steps not shown and / or may omit the steps shown, and the scope of the present invention is not limited in this respect.

[0074] In step 402, the computing device 110 obtains multiple preliminary intention tendencies of the user based on the scalar search results, so as to assign corresponding weights to the multiple preliminary intention tendencies. For example, based on the scalar retrieval results, we obtain (initial intention tendency 1; initial intention tendency 2) and assign weights (initial intention tendency 1, 0.9; initial intention tendency 2, 0.7).

[0075] In step 404 , the computing device 110 obtains a plurality of preliminary intention tendencies of users whose semantic similarity is higher than a predetermined similarity threshold based on the vector retrieval result, and assigns corresponding weights to each of the tendencies.

[0076] For example, based on the vector retrieval results, we obtain (preliminary intention tendency 2; preliminary intention tendency 3, preliminary intention tendency 4), and assign weights (preliminary intention tendency 2, 0.8; preliminary intention tendency 3, 0.7; preliminary intention tendency 4, 0.2).

[0077] In step 406 , the computing device 110 combines the user's preliminary intention tendencies and corresponding weights obtained based on the scalar search result and the vector search result, respectively, to sort the obtained user's preliminary intention tendencies from high to low based on the weights.

[0078] In some embodiments, the user's preliminary intention tendency and corresponding weight obtained by merging the scalar retrieval results and the vector retrieval results include: for the user's current input information, in response to confirming that the same preliminary intention tendency exists in both the scalar retrieval results and the vector retrieval results, excluding the vector retrieval results of the same preliminary intention tendency.

[0079] For example, continuing with the above example, the scalar retrieval results are (preliminary intention tendency 1, 0.9; preliminary intention tendency 2, 0.7); the vector retrieval results are (preliminary intention tendency 2, 0.8; preliminary intention tendency 3, 0.7; preliminary intention tendency 4, 0.2); the two are merged to obtain preliminary intention tendency (preliminary intention tendency 1, 0.9; preliminary intention tendency 2, 0.7; preliminary intention tendency 3, 0.7; preliminary intention tendency 4, 0.2). Preliminary intention tendency 2 appears in both the scalar retrieval and vector retrieval results. At this time, the scalar retrieval results are given priority, and the duplicate vector retrieval results are removed.

[0080] In this way, the characteristics of both scalar retrieval and vector retrieval can be combined. Scalar retrieval provides more accurate retrieval results, and vector retrieval provides closer retrieval results with high similarity in the blind spots of scalar retrieval. The two are integrated to obtain the user's preliminary intention tendency corresponding to the user's current input information.

[0081] Figure 5 The flowchart of the method 500 for determining the current intention of the user and the service type corresponding to the current intention according to an embodiment of the present invention is shown. The method 500 for determining the current intention of the user and the service type corresponding to the current intention can be performed as follows: Figure 1 The computing device 110 shown is executed, and may also be executed on Figure 6It is executed at the electronic device 600. It should be understood that the method 500 for determining the user's current intention and the service type corresponding to the current intention may also include additional steps not shown and / or may omit the steps shown, and the scope of the present invention is not limited in this respect.

[0082] In step 502 , the computing device 110 obtains the user's current input information, the user's historical input information, the user's historical intention information, and the user's preliminary intention tendency ranking to input into the intention recognition model.

[0083] Regarding the intent recognition model, it is, for example, a Large Language Model (LLM), Large Foundation Model (LFM) model, or is constructed based on an LLM model, an LFM model, or a neural network model.

[0084] For example, the user's preliminary intention tendencies are sorted, and the input intention recognition models ranked in the top n, such as the top two preliminary intention tendencies, such as the top three preliminary intention tendencies.

[0085] The user's historical input information may include, for example, the user's input information within t minutes (e.g., t is 10, t is 100), the user's previous m input information (e.g., m is 5, m is 10), and / or the user's historical question and answer information.

[0086] In step 504 , the computing device 110 identifies the user's predicted intention tendency via the intention recognition model, where the predicted intention tendency includes: at least one intention tendency belonging to a predetermined intention tendency set, and / or at least one preliminary intention tendency.

[0087] Regarding the predetermined intent tendency set, for example, FAQ intent (FAQ, Frequently Asked Questions), which is a common intent in this application, it also includes a predetermined intent tendency set for collecting common intents. When the intent recognition model identifies one or more intent tendencies based on the input information and the FAQ intent tendency has the highest probability, the FAQ intent tendency is identified as the user's current intent. If the FAQ intent tendency does not have the highest probability, the identified intent tendencies are ranked.

[0088] For example, the intent recognition model will output at least one predicted intent tendency, for example, 2, and then determine the user's current intent and the business type corresponding to the current intent based on the output predicted intent tendency to further make business recommendations.

[0089] In step 506 , the computing device 110 performs sorting based on the probability of the user's predicted intention tendency to determine the user's current intention and the business type corresponding to the current intention.

[0090] For example, the intent recognition model identifies one or more intents based on input data (the user's current input information, the user's historical input information, the user's historical intent information, and the user's preliminary intention tendency ranking). If an FAQ intent exists and has the highest intent score, the FAQ intent is directly identified as the user's current intent. If there are multiple other intents, the model ranks the intent tendencies, for example, the top two intent tendencies are selected, and then business recommendations are made based on these two intent tendencies. For example, if the business type is 1, the user's intention is to recommend products. The intent recognition model calls the product query interface through FunctionCall and returns a list of product information to the user. If the business type is 2, the user's intention is to recommend applications. The intent recognition model calls the application query interface through FunctionCall and returns a list of applications to the user. If the business type is 3, the user's intention is to recommend promotional activities. The intent recognition model calls the promotional activity query interface through FunctionCall and returns promotional activities to the user. If the business type is 4, the user's intention is to ask questions. The intent recognition model retrieves and enhances the information through the RAG (Retrieval-Augmented Generation) module to generate answers and feeds the answers back to the user. If the business type is empty, the large model determines that if it is a casual chat scenario, it calls the chat module to chat with the user.

[0091] For example, it also includes collecting user input information and intention prediction information of the intention recognition model based on a predetermined period to determine the accuracy of the prediction results of the intention recognition model; providing accuracy feedback information about the intention recognition model to the optimization module to adjust the relevant data stored in the scalar database, vector database and / or predetermined intention tendency set.

[0092] Figure 7 The data flow framework diagram of the intention recognition method according to an embodiment of the present invention is shown. Figure 7 The intention recognition method provided in the embodiment of the present invention is further explained.

[0093] Get current input information from the user; For the current input information, a scalar search is performed through the scalar database to obtain a scalar search result; Vectorize the current input information, then perform vector search through the vector database to obtain vector search results; Combine scalar retrieval results and vector retrieval results to obtain preliminary intent tendency; At least current input information and preliminary intention tendency are input into the intention recognition model (one or more of historical input information and historical intention information may also be input), and the current intention is obtained through the intention recognition model; Based on the current intent and the business field corresponding to the current intent, a recommendation tool is called to make business recommendations, where the recommendation tool includes multiple tools (such as tool a11, tool a21, ..., tool an1); for example, the corresponding recommendation tool is called based on the business type. For example, if the business type corresponding to the current intent is business type a11, tool a11 is called.

[0094] Figure 6 Schematic diagram of an example electronic device 600 that can be used to implement the embodiments of the present specification. Figure 1 The computing device 110 shown can be implemented by an electronic device 600. As shown, the electronic device 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 602 or computer program instructions loaded from a storage unit 608 into a random access memory (RAM) 603. In the random access memory (RAM) 603, various programs and data required for the operation of the electronic device 600 can also be stored. The central processing unit (CPU) 601, the read-only memory (ROM) 602, and the random access memory (RAM) 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0095] Multiple components in the electronic device 600 are connected to the input / output I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, a microphone, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a magnetic disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the electronic device 600 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0096] The various processes and processing described above, such as the method 200 for making service recommendations based on the identified user intent to the method 500 for determining the user's current intent and the service type corresponding to the current intent, can be executed by the central processing unit CPU601. For example, in some embodiments, the method 200 for making service recommendations based on the identified user intent to the method 500 for determining the user's current intent and the service type corresponding to the current intent can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via the read-only memory ROM602 and / or the communication unit 609. When the computer program is loaded into the random access memory RAM603 and executed by the central processing unit CPU601, one or more actions of the method 200 for making service recommendations based on the identified user intent to the method 500 for determining the user's current intent and the service type corresponding to the current intent described above can be performed.

[0097] The present invention relates to methods, apparatuses, systems, electronic devices, computer-readable storage media and / or computer program products. The computer program products may include computer-readable program instructions for executing various aspects of the present invention.

[0098] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or raised-in-groove structure on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0099] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge computing devices. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0100] The computer program instructions for performing the operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, the state information of the computer-readable program instructions is used to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), so that the electronic circuit can execute the computer-readable program instructions, thereby implementing various aspects of the present invention.

[0101] Various aspects of the present invention are described herein with reference to flowcharts and / or step diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each step of the flowcharts and / or step diagrams, and any combination of the steps in the flowcharts and / or step diagrams, can be implemented by computer-readable program instructions.

[0102] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine such that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more steps in the flowchart and / or step diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more steps in the flowchart and / or step diagram.

[0103] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more steps in the flowchart and / or step diagram.

[0104] The flowcharts and step diagrams in the accompanying drawings show the possible architectures, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each step in the flowchart or step diagram can represent a module, program segment or part of an instruction, and a module, program segment or part of an instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the steps can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each step in the step diagram and / or flowchart, and the combination of the steps in the step diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0105] While various embodiments of the present invention have been described above, the above descriptions are intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for recommending services based on identified user intentions, characterized in that: include: Based on the user's current input information, scalar search and vector search are performed in the keyword database respectively; Sort the scalar retrieval results and vector retrieval results to obtain the user's preliminary intention tendency ranking; At least sorting the user's current input information and the user's preliminary intention tendency, and inputting them into an intention recognition model, so as to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model; The intention recognition model is constructed based on a neural network model; as well as Based on the service type corresponding to the current intention of the user, a recommendation tool corresponding to the service type is called to make service recommendations based on the current intention of the user.

2. The method according to claim 1, characterized in that The business types include: product recommendation, application recommendation, activity recommendation, question inquiry and / or chat.

3. The method according to claim 1, wherein The keyword database includes: a scalar database constructed based on scalar data regarding product data, activity data, application data, and / or problem data corresponding to a business type; and A vector database is constructed based on vectorized commodity data, activity data, application data and / or problem data corresponding to a business type.

4. The method according to claim 1, wherein Performing scalar and vector searches in a keyword database includes: Vectorizing the user's current input information to obtain vectorized current input information; Performing similarity search in a vector database for the vectorized current input information; and Based on the semantic similarity, at least one business type and at least one preliminary intention tendency corresponding to the user's current input information are identified.

5. The method according to claim 1, characterized in that Sorting based on scalar retrieval results and vector retrieval results to obtain the user's preliminary intention tendency ranking includes: Based on the scalar search results, multiple preliminary intention tendencies of the user are obtained, so as to assign corresponding weights to the multiple preliminary intention tendencies; Based on the vector search results, obtaining multiple preliminary intention tendencies for users whose semantic similarity is higher than a predetermined similarity threshold, and assigning corresponding weights to each of them; and The user's preliminary intention tendencies and corresponding weights obtained based on the scalar retrieval results and the vector retrieval results are merged to sort the obtained user's preliminary intention tendencies from high to low based on the weights.

6. The method according to claim 5, characterized in that The user's preliminary intention and corresponding weight obtained by combining the scalar search results and vector search results include: In response to confirming that both a scalar search result and a vector search result for the same preliminary intention tendency exist for the user's current input information, the vector search results for the same preliminary intention tendency are excluded.

7. The method according to claim 1, characterized in that At least sorting the user's current input information and the user's preliminary intention tendency and inputting them into the intention recognition model to determine the user's current intention and the business type corresponding to the current intention through the intention recognition model includes: Obtain the user's current input information, the user's historical input information, the user's historical intention information, and the user's preliminary intention tendency ranking to input into the intention recognition model; identifying, via the intention recognition model, a predicted intention tendency of the user, the predicted intention tendency including: at least one intention tendency belonging to a predetermined intention tendency set, and / or at least one preliminary intention tendency; and Sorting is performed based on the probability of the user's predicted intention tendency to determine the user's current intention and the business type corresponding to the current intention.

8. A computing device, characterized in that include: at least one processing unit; At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the apparatus to perform the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a machine, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a machine, the method according to any one of claims 1 to 7 is performed.

Images