CN115907892A - User intention identification method and device based on user multi-interest evolution - Google Patents
User intention identification method and device based on user multi-interest evolution Download PDFInfo
- Publication number
- CN115907892A CN115907892A CN202211249675.0A CN202211249675A CN115907892A CN 115907892 A CN115907892 A CN 115907892A CN 202211249675 A CN202211249675 A CN 202211249675A CN 115907892 A CN115907892 A CN 115907892A
- Authority
- CN
- China
- Prior art keywords
- user
- intention
- interest
- content
- commodities
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The method comprises the steps of constructing a content intention pool, and selecting content intentions reaching a preset user interest probability from the content intention pool according to a historical interaction sequence of a user; judging whether the historical commodities are related commodities with content intentions or not by utilizing the correlation judging network to obtain an intention chain consisting of the related commodities, and obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities; and selecting an intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user. The invention can enable the user to enjoy better and personalized platform service, thereby improving the user experience; the user's purchasing intention is predicted more carefully, so that the user can be deeply understood, and the user's adhesion is increased; the business amount is increased by actively adjusting the strategy in time to promote the user to reach the transaction.
Description
Technical Field
The application relates to the technical field of e-commerce platforms, in particular to a user intention identification method and device based on multi-interest evolution of users.
Background
With the development of the internet, e-commerce platforms have been greatly developed, which serve hundreds of millions of users and have billions of goods. The user can find the psychographic commodity through two services of personalized search and recommendation provided by the platform, so that the transaction is finally completed. In order to enhance the user experience and provide more personalized services to the user, the platform has made a great effort in understanding the user's intent.
User intent understanding can be divided into two levels: one is content-based understanding, namely: which goods or which searches may be more interesting to the user; the other is purpose-based intent understanding, i.e., determining whether the user's current intent is for purchase or browsing. For most of the existing works, understanding of the user's intended purpose is often ignored, however, it is very valuable to accurately understand the user's current intended purpose, whether it is for the user or the seller.
It is therefore necessary for the platform to make some proactive policy adjustments appropriately according to the user's real-time intent in the search scenario.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a method and an apparatus for identifying a user intention based on a multi-interest evolution of a user, so as to solve or partially solve the above technical problems.
In view of the above, a first aspect of the present application provides a user intention identification method based on user multi-interest evolution, including:
constructing a content intention pool, and selecting content intentions reaching a preset user interest probability from the content intention pool according to a historical interaction sequence of a user;
judging whether the historical commodities are related commodities with content intentions or not by utilizing the relevance judgment network to obtain an intention chain consisting of the related commodities, and obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and selecting an intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
As a preferred scheme of the user intention identification method based on the multi-interest evolution of the user, a given user history interaction sequence S = { e = { (e) } 1 ，e 2 ，…，e T Regarding the sequence as environmental stimulus, and acquiring the correlation alpha of the ith commodity and the jth primary interest of the sequence i j ：
In the formula, e i An expression of the ith good that is a historical sequence of interactions;
in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the jth primary interest by the historical interaction sequence S:
as a preferred scheme of a user intention identification method based on multi-interest evolution of a user, constructing high-level interest information by using the representation of the obtained historical interaction sequence S on the jth primary interest;
given a content intent set P = { P) to be learned 1 ，p 2 ，...p |P| }, representing prior knowledge experience; where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula, beta i j A relevance of the ith primary interest to the content intent j; σ () is a sigmod function.
As a preferred scheme of the user intention identification method based on the multi-interest evolution of the user, the probability that the historical commodity is divided into the jth interest chain is calculated by utilizing the constructed high-level interest information:
wherein, MLP (e) i ，c j ) A function for measuring the contribution degree of a commodity i to a high-level interest j;
and carrying out fine-grained interest evolution under different high-level interests according to the interest chain.
As a preferred scheme of a user intention identification method based on user multi-interest evolution, a time sequence depth model is based and an interest state z at the moment t is obtained j，t State at time t +1
Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+1 ，e t+1 ]))
f t，j ＝σ f (MLP f ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t,j ；f t,j ；r t,j Respectively representing three gating settings, a t+1 The operation behavior of the user at the t +1 th moment; e.g. of the type t+1 Is the expression of the t +1 st commodity;
and fusing the evolved high-level interests and the search terms of the user to optimize the model.
A second aspect of the present application provides a user intention recognition apparatus based on user multi-interest evolution, including:
the multi-intention extraction module is used for constructing a content intention pool, and selecting content intentions reaching the preset user interest probability from the content intention pool according to the historical interaction sequence of the user;
the multi-purpose chain generating module is used for judging whether the historical commodities are related commodities with content purposes by utilizing the correlation judging network to obtain purpose chains consisting of the related commodities;
the multi-intention evolution module is used for obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and the multi-intention management module is used for selecting the intention expression with the correlation reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
As a preferred scheme of the user intention identification device based on the evolution of the multiple interests of the user, the multiple intention extraction module comprises a bottom-up submodule:
the bottom-up submodule is used for enabling a given user history interaction sequence S = { e } 1 ，e 2 ，···，e T Consider as environmental stimulus, obtain the correlation α of the ith commodity and the jth primary interest in the sequence i j ：
In the formula, e i An expression of the ith good that is a historical sequence of interactions;
in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the jth primary interest by the historical interaction sequence S:
as a preferred scheme of the user intention identification device based on the evolution of the multiple interests of the user, the multiple intention extraction module comprises top-down sub-modules:
the top-down sub-module is used for constructing high-level interest information by using the obtained characterization of the j-th primary interest by the historical interaction sequence S;
in the top-down sub-module, a content meaning set P = { P) to be learned is given 1 ,p 2 ,…p |P| ) Representing prior knowledge and experience; where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula, beta i j A relevance of the ith primary interest to the content intent j; σ () is a sigmod function.
As a preferred scheme of the user intention identification device based on the evolution of the multiple interests of the user, the multiple intention extraction module comprises an interest chain analysis submodule:
the interest chain analysis submodule is used for calculating the probability that the historical commodities are divided into the jth interest chain by utilizing the constructed high-level interest information:
wherein, MLP (e) i ，c j ) Is a function for measuring the contribution degree of the commodity i to the high-level interest j;
and carrying out fine-grained interest evolution under different high-level interests according to the interest chain.
As a preferred scheme of the user intention identification device based on the evolution of multiple interests of the user, the time sequence depth model is based on the interest state z at the time t j，t State at time t +1
Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+1 ，e t+1 ]))
f t，j ＝σ f (MLP f ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t，j ；f t，j ；r t，j Respectively representing three gating settings, a t+1 The operation behavior of the user at the t +1 th moment; e.g. of the type t+1 Is the expression of the t +1 st commodity;
the system also comprises a model optimization module used for fusing the evolved high-level interests and the search terms of the user and optimizing the model.
A third aspect of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the method for identifying a user intention based on user multi-interest evolution according to the first aspect is implemented.
A fourth aspect of the present application proposes a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute a method for identifying a user's intention based on user multi-interest evolution according to the first aspect.
From the above description, according to the technical scheme provided by the application, the content intention reaching the preset user interest probability is selected from the content intention pool by constructing the content intention pool and according to the historical interaction sequence of the user; judging whether the historical commodities are related commodities with content intentions or not by utilizing the correlation judging network to obtain an intention chain consisting of the related commodities, and obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities; and selecting an intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user. The invention can enable the user to enjoy better and personalized platform service, thereby improving the user experience; the user can be deeply understood by more carefully predicting the purchasing intention of the user, so that the adhesion of the user is increased; the business amount is increased by actively adjusting the strategy in time to promote the user to reach the transaction.
Drawings
In order to more clearly illustrate the technical solutions in the present application or the related art, the drawings needed to be used in the description of the embodiments or the related art will be briefly introduced below, and it is obvious that the drawings in the following description are only embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a schematic flow chart of a user intention identification method based on user multi-interest evolution according to an embodiment of the present application;
FIG. 2 is a schematic diagram of an evolution model related to a user intention identification method based on user multi-interest evolution according to an embodiment of the application;
FIG. 3 is a schematic diagram of an interest evolution network of a user intention identification method based on user multi-interest evolution according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a user intention recognition apparatus based on user multi-interest evolution according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is further described in detail below with reference to the accompanying drawings in combination with specific embodiments.
It should be noted that technical terms or scientific terms used in the embodiments of the present application should have a general meaning as understood by those having ordinary skill in the art to which the present application belongs, unless otherwise defined. The use of the terms "comprising" or "including" and the like in the embodiments of the present application, means that the element or item appearing before the term covers the element or item listed after the term and its equivalents, without excluding other elements or items.
In the prior art, the user needs using the search function are often quite clear, and the platform only needs to recommend personalized related goods to the user according to the user interests embodied in the search terms, such as: a user searches for the search term "dress", and then the platform only needs to continuously recommend to the user the relevant items of "dress" that he may be interested in. However, the investigation on the users of the electronic commerce finds that the purpose of most users using the search function is only to randomly visit and stroll, and the user does not have a very specific requirement. It is therefore necessary for the platform to make some proactive policy adjustments appropriately according to the user's real-time intent in the search scenario.
In the related art, there are simple examples of active policy adjustment, for example, if a user currently searches for a search word "one-piece dress", and if the user has a purchase intention under the search word, the platform may further recommend some more detailed search words to the user appropriately, explore more fine-grained interest preferences of the user, or the platform may issue some time-limited coupons to the user to stimulate the user to complete consumption. If the user does not have the purchasing intention under the search word, the platform can recommend some more divergent search words to the user appropriately to explore other aspects that the user may be interested in, or stimulate the user to consume by creating a supporting product bundle.
Given a user's historical behavior log and current search terms, it is determined whether the user has a purchase intention under the current search terms. Due to the randomness and diversity of user behavior, the user's intent for multiple different categories may be discovered from the user's behavior history. That is, a complete sequence of user actions can actually be split into multiple sub-sequences of actions for different categories of intent, where each sub-sequence reflects the intent of the user for a particular category. Using only one vector to capture a user's intent tends to impair the ability to model a single intent, limiting the performance of the model. Therefore, in recent years, attention has been focused on the use of the multi-intent modeling method. However, such methods only emphasize the multiple intentions of the user extracted from the sequence, and neglect the evolution process of the user's intentions on the multiple intentions.
Taking the historical sequence of a certain user as an example, 3 different types of intentions were found, namely T-shirts, bags and skirts. Taking a T-shirt as an example, a user clicks a first T-shirt and then purchases a second T-shirt, intuitively, and the purchasing intention of the user on the T-shirt is continuously strengthened in the process. However, after a while, the user purchases a T-shirt again, the user's intention to purchase the T-shirt is drastically reduced because the user's intention to purchase the T-shirt is satisfied. This process of intent variation over time and user operational behavior is intent evolution. The intention evolution modeling mode is logical, and from the psychological point of view, the occurrence of the purchasing behavior of the user is a natural sequence process, the user firstly has the interested content intention, then browses and compares the related commodities under the specific content intention, and finally purchases occur.
One important challenge faced by the intent modeling process is how to accurately extract multiple user content intents from a sequence. The methods for extracting a plurality of content intents of a user from a sequence mainly comprise two types, namely an implicit multi-content intention mining method and a displayed multi-content intention mining method. Implicit methods are mainly by self-attention, however such methods do not promote significantly compared to the method of the simple drawing; the explicit method requires learning an expression in each category interacted by the user, and has poor expandability.
In order to accurately understand the current intention of the user, it is necessary to make active policy adjustments appropriately according to the real-time intention of the user in different search scenarios. In view of this, referring to fig. 1, fig. 2 and fig. 3, an embodiment of the present invention provides a user intention identification method based on user multi-interest evolution, including the following steps:
s1, constructing a content intention pool, and selecting content intentions reaching a preset user interest probability from the content intention pool according to a historical interaction sequence of a user;
s2, judging whether the historical commodities are related commodities with content intentions or not by utilizing the correlation judgment network to obtain an intention chain consisting of the related commodities;
s3, obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and S4, selecting the intention expression with the correlation reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
In this embodiment, U = { U = is used 1 ,u 2 ,…u |U| Represents all users, I = { I = } 1 ,i 2 ,…i |I| Represents all goods, Q = { Q = } 1 ,q 2 ,…q |Q| And represents all search terms, wherein | U |, | I | and | Q | represent the number of all users, all commodities, and all search terms, respectively.
In this example, Z i And Z Q Representing other sets of auxiliary information relating to the articles and search terms, respectively, including category information, tag information, term information, etc., e.g. corresponding to the jth article i in the sequence j And search term q j The auxiliary information is
And &>
A = { a1, a2, a3, a4} represents different operation behaviors of the user, and the physical meaning is as follows: click, buy, and collect.
For each user u use:
The task of this embodiment is the prediction of the purchasing intent under the current search term, i.e. given the historical interaction sequence Su of the user u and the information of the search term at the user's current time
And judging whether the user can buy under the current search word.
In order to accurately model multiple interests under the user complex cognitive classification standard, namely, extract multiple interests meeting the user cognitive classification standard, considering the difference between the user complex cognitive classification standard and the category information, the forming process of the cognitive classification standard is involved. In a real scene, a user gradually forms a complex cognitive classification standard in the process of continuously browsing and comparing related commodities. The cognitive forming process of the human is similar to the perceptual forming process of the human, and the cognitive classification standard forming process in the consuming process of the user is modeled by simulating the perceptual forming process of the human in the cognitive psychology in daily life, wherein the perceptual forming process is as follows (a perceptual model in the cognitive psychology).
For this purpose, perceptual models in cognitive psychology are introduced. By taking the historical interaction sequence of the user as environmental stimulus, on the basis of the existing cognitive classification standard of an individual, a plurality of different cognitive interests of the user are extracted through bottom-up processing and top-down processing, and corresponding interest chains are split.
In particular, the method comprises the following steps of, given user history interaction sequence S = { e = 1 ，e 2 ，···，e T Regarding as environmental stimulus, obtain the correlation α of ith commodity and jth primary interest i j ：
In the formula, e i An expression of the ith good that is a historical sequence of interactions; in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the jth primary interest by the historical interaction sequence S:
after the representation of the primary interest is constructed, the representation of the primary interest is further selected and integrated, so that high-level interest information is constructed, and the high-level perception interest finally determines the behavior of a user. Constructing high-level interest information by simulating a learning process of high-level cognition and utilizing the obtained representations of the historical interaction sequence S on k different primary interests;
given a content intent set P = { P) to be learned 1 ,p 2 ,…p |P| Represents the prior experience knowledge of the user. Where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula, beta i j The contribution of the ith primary interest to the high level interest j; σ () is a sigmod function.
In this embodiment, the probability that the historical commodity is divided into the jth interest chain is calculated by using the constructed high-level interest information:
wherein, MLP (e) i ，c j ) The contribution degree of the commodity i to the high-level interest j;
according to the interest chain, more fine-grained interest evolution can be performed under different high-level interests.
To assist FIG. 3, in this embodiment, in order toThe influence of different operation behaviors of the user on the interest evolution can be modeled, and an interest evolution network is constructed. Based on a time sequence depth model, according to the interest state z at the moment t j，t State z at time t +1 j，t+1 Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+1 ，e t+1 ]))
f t，j ＝σ f (MLP f ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t，j ；f t，j ；r t，j Respectively representing three gating settings, a t+1 The user operation behavior at the t +1 th moment; e.g. of the type t+1 Is the expression of the t +1 st commodity;
and fusing the evolved high-level interests and the search terms of the user, and optimizing the evolution model.
Because the embodiment of the invention evolves based on the intention chain, when the historical commodity e t+1 As high-level interest c j When the goods are truly related, q j|i =1, satisfying:
otherwise, q j|i ＝0，z j，t+1 ＝z j，t+1 . After the high-level interest evolution, the final j high-level interest z k ＝z k，T (ii) a Obtaining accurate user multi-interest Z = { Z = { Z) from the method 1 ，z 2 ，…，z k }。
In this embodiment, the evolved high-level interests and the search terms of the user are fused, and the evolution model is optimized. The fusion formula is as follows:
wherein h is f Extracting the information most relevant to the current query word; finally, calculating the final score through a two-layer MLP network:
based on this score, the final loss function is as follows:
in summary, the content intention reaching the preset user interest probability is selected from the content intention pool according to the historical interaction sequence of the user by constructing the content intention pool; judging whether the historical commodities are related commodities with content intentions or not by utilizing the relevance judgment network to obtain an intention chain consisting of the related commodities, and obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities; and selecting an intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user. The invention can enable the user to enjoy better and personalized platform service, thereby improving the user experience; the user's purchasing intention is predicted more carefully, so that the user can be deeply understood, and the user's adhesion is increased; the business amount is increased by actively adjusting the strategy in time to promote the user to reach the transaction.
It should be noted that the method of the embodiment of the present application may be executed by a single device, such as a computer or a server. The method of the embodiment can also be applied to a distributed scene and completed by the mutual cooperation of a plurality of devices. In this distributed scenario, one device of the multiple devices may only perform one or more steps of the method of the embodiment of the present application, and the multiple devices interact with each other to complete the method.
It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
Referring to fig. 4, based on the same inventive concept, corresponding to any of the above-mentioned embodiments, the present application further provides a user intention identification apparatus based on multi-interest evolution of a user, including:
the multi-intention extraction module 1 is used for constructing a content intention pool, and selecting content intentions reaching a preset user interest probability from the content intention pool according to a historical interaction sequence of a user;
the multi-purpose chain generating module 2 is used for judging whether the historical commodities are related commodities with content purposes by utilizing the relevance judging network to obtain purpose chains consisting of the related commodities;
the multi-intention evolution module 3 is used for obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and the multi-intention management module 4 is used for selecting the intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
In this embodiment, the multi-intent extraction module 1 includes a bottom-up sub-module 11:
a bottom-up submodule 11 for convertingGiven user history interaction sequence S = { e = 1 ，e 2 ，···，e T Regarding the sequence as environmental stimulus, and acquiring the correlation alpha of the ith commodity and the jth primary interest of the sequence i j ：
In the formula, e i An expression of the ith good that is a historical sequence of interactions;
in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the jth primary interest by the historical interaction sequence S:
/>
in this embodiment, the multi-intent extraction module 1 includes top-down sub-modules 12:
the top-down sub-module 12 is used for constructing high-level interest information by using the obtained characterization of the j-th primary interest by the historical interaction sequence S;
in the top-down submodule 12, a content ideogram set P = { P } to be learned is given 1 ,p 2 ,…p |P| Represents prior knowledge and experience; where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula, beta i j A relevance of the ith primary interest to the content intent j; σ () is a sigmod function.
In this embodiment, the multi-intent extraction module includes an interest chain analysis sub-module 13:
the interest chain analysis submodule 13 is configured to calculate, by using the constructed high-level interest information, a probability that the historical commodity is divided into a jth interest chain:
wherein, MLP (e) i ,c j ) The contribution degree of the commodity i to the high-level interest j;
according to the interest chain, more fine-grained interest evolution can be performed under different high-level interests.
In this embodiment, in the multi-intent evolution module 3, based on a time-series depth model, the interest state z at time t is determined j,t State z at time t +1 j,t+1 Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+I ，e t+1 ]))
f t，j ＝σ f (MLP f ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t，j ；f t，j ；r t，j Respectively representing three gating settings, a t+1 The operation behavior of the user at the t +1 th moment; e.g. of the type t+1 Is the expression of the t +1 st commercial product.
In this embodiment, the system further includes a model optimization module 5, which is used for fusing the evolved high-level interests and the search terms of the user to optimize the model.
For convenience of description, the above system is described with the functions divided into various modules, which are described separately. Of course, the functionality of the various modules may be implemented in the same one or more pieces of software and/or hardware in the practice of the present application.
The device of the above embodiment is used for implementing the user intention identification method based on the multi-interest evolution of the user in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.
Based on the same inventive concept, corresponding to the method of any embodiment described above, the present application further provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the method for identifying a user intention based on the evolution of multiple interests of the user according to any embodiment described above is implemented.
Fig. 5 is a schematic diagram illustrating a more specific hardware structure of an electronic device according to this embodiment, where the electronic device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 are communicatively coupled to each other within the device via bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solutions provided in the embodiments of the present disclosure.
The Memory 1020 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present specification is implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called to be executed by the processor 1010.
The input/output interface 1030 is used for connecting an input/output module to input and output information. The i/o module may be configured as a component within the device (not shown) or may be external to the device to provide corresponding functionality. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.
The communication interface 1040 is used for connecting a communication module (not shown in the drawings) to implement communication interaction between the present apparatus and other apparatuses. The communication module can realize communication in a wired mode (such as USB, network cable and the like) and also can realize communication in a wireless mode (such as mobile network, WIFI, bluetooth and the like).
The bus 1050 includes a path to transfer information between various components of the device, such as the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040.
It should be noted that although the above-mentioned device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will appreciate that the above-described apparatus may also include only those components necessary to implement the embodiments of the present description, and not necessarily all of the components shown in the figures.
The electronic device of the above embodiment is used for implementing the user intention identification method based on the multi-interest evolution of the user in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.
Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application further provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the user intent identification method based on the evolution of multiple interests of a user as described in any of the above embodiments.
Computer-readable media, including both permanent and non-permanent, removable and non-removable media, for storing information may be implemented in any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
The computer instructions stored in the storage medium of the foregoing embodiment are used to enable the computer to execute the method for identifying user intention based on user multiple interest evolution described in any embodiment above, and have the beneficial effects of the corresponding method embodiments, which are not described herein again.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the context of the present application, technical features in the above embodiments or in different embodiments may also be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application described above, which are not provided in detail for the sake of brevity.
In addition, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown in the provided figures for simplicity of illustration and discussion, and so as not to obscure the embodiments of the application. Furthermore, devices may be shown in block diagram form in order to avoid obscuring embodiments of the application, and this also takes into account the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the embodiments of the application are to be implemented (i.e., specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the application, it should be apparent to one skilled in the art that the embodiments of the application can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative instead of restrictive.
While the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures, such as Dynamic RAM (DRAM), may use the discussed embodiments.
The present embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present application are intended to be included within the scope of the present application.
Claims (10)
1. The user intention identification method based on the multi-interest evolution of the user comprises the following steps:
constructing a content intention pool, and selecting content intentions reaching a preset user interest probability from the content intention pool according to a historical interaction sequence of a user;
judging whether the historical commodities are related commodities with content intentions or not by utilizing the relevance judgment network to obtain an intention chain consisting of the related commodities, and obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and selecting an intention expression with the relevance reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
2. The user intent identification method based on user multi-interest evolution as claimed in claim 1, wherein a given user history interaction sequence S = { e = { e = 1 ，e 2 ，…，e T Regarding the sequence as environmental stimulus, acquiring the ith commodity and the jth commodity in the sequenceRelevance of primary interest alpha i j ：
In the formula, e i An expression of the ith good that is a historical sequence of interactions;
in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the history interaction sequence S on the jth primary interest:
3. the method for identifying the user intention evolved based on the multiple interests of the user as claimed in claim 2, wherein the obtained historical interaction sequence S is used for characterizing the jth primary interest to construct high-level interest information;
given a set of content views to be learned P = { P = { P = 1 ,p 2 ,…p |P| }, representing prior knowledge experience; where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula (I), the compound is shown in the specification,
a relevance of the ith primary interest to the content intent j; σ () is a sigmod function.
4. The user intention recognition method based on the multi-interest evolution of the users as claimed in claim 3, wherein the probability that the historical commodities are divided into the jth interest chain is calculated by using the constructed high-level interest information:
wherein, MLP (e) i ，c j ) Is a function for measuring the contribution degree of the commodity i to the high-level interest j;
and carrying out fine-grained interest evolution under different high-level interests according to the interest chain.
5. The method for identifying user intention based on evolution of multiple interests of user as claimed in claim 4, wherein the interest state z at t moment is based on a time sequence depth model j，t State at time t +1
Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+1 ，e t+1 ]))
f t，j ＝σ f (MLP F ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t，j ；f t，j ；r t，j Respectively representing three gating settings, a t+1 The operation behavior of the user at the t +1 th moment; e.g. of the type t+1 For the t +1 st articleExpressing;
and fusing the evolved high-level interests and the search terms of the user to optimize the model.
6. A user intention recognition device based on multi-interest evolution of a user comprises:
the multi-intention extraction module is used for constructing a content intention pool, and selecting content intentions reaching the preset user interest probability from the content intention pool according to the historical interaction sequence of the user;
the multi-purpose chain generating module is used for judging whether the historical commodities are related commodities with content purposes by utilizing the correlation judging network to obtain purpose chains consisting of the related commodities;
the multi-intention evolution module is used for obtaining the final intention expression of the user according to the intention chain and the operation behaviors of the user on different commodities;
and the multi-intention management module is used for selecting the intention expression with the correlation reaching a preset value according to the current search word of the user so as to predict the purchasing intention of the user.
7. The apparatus of claim 6, wherein the multi-intent extraction module comprises a bottom-up sub-module:
the bottom-up submodule is used for enabling a given user history interaction sequence S = { e } 1 ，e 2 ，…，e T Regarding the sequence as environmental stimulus, and acquiring the correlation alpha of the ith commodity and the jth primary interest of the sequence i j ：
In the formula, e i An expression of the ith commodity which is a historical interaction sequence;
in obtaining an article e i And after the correlation with the jth primary interest, aggregating all T commodities to obtain the characterization of the history interaction sequence S on the jth primary interest:
8. the apparatus for identifying user intent based on evolution of multiple interests of the user as claimed in claim 7, wherein said multiple intent extraction module comprises top-down sub-modules:
the top-down sub-module is used for constructing high-level interest information by utilizing the obtained characterization of the history interaction sequence S on the jth primary interest;
in the top-down sub-module, a content meaning set P = { P) to be learned is given 1 ，p 2 ，...p |P| Represents prior knowledge and experience; where | P | is a given fixed quantity, the high-level interest information is represented as:
in the formula (I), the compound is shown in the specification,
a relevance of the ith primary interest to the content intent j; σ () is a sigmod function.
9. The apparatus for identifying user's intention evolving based on multi-user interest as claimed in claim 8, wherein the multi-intention extracting module includes an interest chain analyzing sub-module:
the interest chain analysis submodule is used for calculating the probability that the historical commodities are divided into the jth interest chain by utilizing the constructed high-level interest information:
wherein, MLP (e) i ，c j ) Is a function for measuring the contribution degree of the commodity i to the high-level interest j;
and carrying out fine-grained interest evolution under different high-level interests according to the interest chain.
10. The apparatus for identifying user's intention based on evolution of multiple interests of user as claimed in claim 9, wherein said multiple intention evolution module is based on time sequence depth model and according to interest state z at t moment j，t State at time t +1
Comprises the following steps:
g t，j ＝σ g (MLP g ([z j，t ，a t+1 ，e t+1 ]))
f t，j ＝σ f (MLP f ([z j，t ，a t+1 ，e t+1 ]))
r t，j ＝σ r (MLP r ([z j，t ，a t+1 ，e t+1 ]))
in the formula, g t，j ；f t，j ；r t，j Respectively representing three gating settings, a t+1 The operation behavior of the user at the t +1 th moment; e.g. of the type t+1 Is the expression of the t +1 st commodity;
the system also comprises a model optimization module used for fusing the evolved high-level interests and the search terms of the user and optimizing the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211249675.0A CN115907892A (en) | 2022-10-12 | 2022-10-12 | User intention identification method and device based on user multi-interest evolution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211249675.0A CN115907892A (en) | 2022-10-12 | 2022-10-12 | User intention identification method and device based on user multi-interest evolution |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115907892A true CN115907892A (en) | 2023-04-04 |
Family
ID=86484080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211249675.0A Pending CN115907892A (en) | 2022-10-12 | 2022-10-12 | User intention identification method and device based on user multi-interest evolution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115907892A (en) |
-
2022
- 2022-10-12 CN CN202211249675.0A patent/CN115907892A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784455B (en) | Article recommendation method and recommendation equipment | |
| JP6719727B2 (en) | Purchase behavior analysis device and program | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| CN111815415A (en) | Commodity recommendation method, system and equipment | |
| CN111242729A (en) | Serialization recommendation method based on long-term and short-term interests | |
| CN108230009B (en) | User preference prediction method and device and electronic equipment | |
| TW201520790A (en) | Individualized data search | |
| US11887184B2 (en) | Techniques to facilitate online commerce by leveraging user activity | |
| CN113781149A (en) | Information recommendation method and device, computer-readable storage medium and electronic equipment | |
| CN111680213B (en) | Information recommendation method, data processing method and device | |
| CN111310035A (en) | Recommendation method and device based on psychological and behavioral characteristics | |
| CN111026973B (en) | Commodity interest degree prediction method and device and electronic equipment | |
| CN112036987B (en) | Method and device for determining recommended commodity | |
| CN110851694A (en) | Personalized recommendation system based on user memory network and tree structure depth model | |
| CN116910357A (en) | Data processing method and related device | |
| CN111967924A (en) | Commodity recommendation method, commodity recommendation device, computer device, and medium | |
| CN116204709A (en) | Data processing method and related device | |
| WO2023050143A1 (en) | Recommendation model training method and apparatus | |
| CN115907892A (en) | User intention identification method and device based on user multi-interest evolution | |
| CN114971760A (en) | Vehicle type recommendation method and device based on big data, electronic equipment and medium | |
| CN116764236A (en) | Game prop recommending method, game prop recommending device, computer equipment and storage medium | |
| CN113947431A (en) | User behavior quality evaluation method, device, equipment and storage medium | |
| CN114429384A (en) | Intelligent product recommendation method and system based on e-commerce platform | |
| CN113792952A (en) | Method and apparatus for generating a model | |
| JP2020154464A (en) | Estimation device, estimation method and estimation program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |