Behavior sequence-based test question recommendation method and device, equipment and storage medium

Abstract

The application belongs to the technical field of Internet application, and discloses a test question recommendation method and device based on behavior sequence, equipment and storage medium. The method comprises the following steps: determining the behavior sequence similarity according to the behavior sequence of a target user and the behavior sequence of a candidate user; determining the user similarity according to the behavior sequence similarity; determining the similar users of the target user in the candidate users according to the preset selection quantity and the user similarity, and obtaining the wrong questions of the similar users in a preset time range as candidate recommended test questions; obtaining the association weight data between the historical wrong questions of the target user and the candidate recommended test questions, determining the test question recommendation strategy according to the association weight data and the behavior sequence of the target user; determining the target recommended test question in the candidate recommended test questions according to the test question recommendation strategy, and recommending the target recommended test question to the target user. Through the above method, the user characteristics are fully mined, the personalized test question recommendation is provided, the characteristic change is dynamically updated, and the flexibility is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet application, and particularly relates to a question recommendation method and device based on behavior sequence, equipment and storage medium. BACKGROUND

[0002] Nowadays, online exercises have become a social trend and are widely used, for example, a series of competition activities, which have given rise to a large number of online exercise training systems on the market, so that users can start exercises at any time and anywhere. However, these training systems have many limitations and limitations. The user portrait is not constant, and the user's ability level will change during the exercise process. Most online test systems do not dig deeply enough into the user portrait, and the test question recommendation cannot be flexibly adapted or has strong hysteresis.

[0003] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0004] The main purpose of the present application is to provide a question recommendation method and device based on behavior sequence, equipment and storage medium, which aims to solve the technical problems that most online test systems do not dig deeply enough into the user portrait, and the test question recommendation cannot be flexibly adapted or has strong hysteresis in the prior art.

[0005] To achieve the above purpose, the present application provides a question recommendation method based on behavior sequence, which comprises the following steps:

[0006] Obtain the behavior sequence of the target user and the behavior sequence of the candidate user, and determine the behavior sequence similarity between the target user and the candidate user according to the behavior sequence of the target user and the behavior sequence of the candidate user;

[0007] According to the behavior sequence similarity, determine the user similarity between the target user and the candidate user;

[0008] According to the preset selection quantity and the user similarity, determine the similar user of the target user in the candidate user, and obtain the wrong questions of the similar user in the preset time range as the candidate recommended test questions;

[0009] Obtain the association weight data between the target user's historical wrong questions and the candidate recommended test questions, and determine the test question recommendation strategy according to the association weight data and the behavior sequence of the target user;

[0010] According to the test question recommendation strategy, determine the target recommended test question in the candidate recommended test question, and recommend the target recommended test question to the target user.

[0011] Optionally, the behavior sequence comprises a plurality of state string data arranged in a preset time sequence, the state string data comprising access data and operation data corresponding to the access data, and the determining the behavior sequence similarity between the target user and the candidate user according to the behavior sequence of the target user and the behavior sequence of the candidate user comprises:

[0012] connecting the state string data in the behavior sequence of the target user in sequence to obtain a behavior state sequence of the target user;

[0013] connecting the state string data in the behavior sequence of the candidate user in sequence to obtain a behavior state sequence of the candidate user;

[0014] determining the behavior state order similarity, the behavior state transition similarity and the behavior state value similarity between the target user and the candidate user according to the behavior state sequence of the target user and the behavior state sequence of the candidate user;

[0015] determining the behavior sequence similarity between the target user and the candidate user according to the behavior state order similarity, the behavior state transition similarity and the behavior state value similarity.

[0016] Optionally, the determining the behavior state order similarity, the behavior state transition similarity and the behavior state value similarity between the target user and the candidate user according to the behavior state sequence of the target user and the behavior state sequence of the candidate user comprises:

[0017] determining a maximum common state sub-sequence and a maximum common state sub-sequence length corresponding to the maximum common state sub-sequence according to the behavior state sequence of the target user and the behavior state sequence of the candidate user;

[0018] determining a common state transition number, a state transition and number, a state string and number and a common state string number according to the behavior state sequence of the target user and the behavior state sequence of the candidate user;

[0019] determining the behavior state order similarity according to a corresponding relationship between the maximum common state sub-sequence length, the state string and number and the behavior state order similarity, the maximum common state sub-sequence length and the state string and number;

[0020] determining the behavior state transition similarity according to a corresponding relationship between the common state transition number, the state transition and number and the behavior state transition similarity, the common state transition number and the state transition and number;

[0021] According to the state string and number, the correspondence between the common state string number and the behavior state value similarity, the state string and number, and the common state string number, determine the behavior state value similarity.

[0022] Optionally, the determining the behavior sequence similarity between the target user and the candidate user according to the behavior state order similarity, the behavior state transition similarity, and the behavior state value similarity comprises:

[0023] Obtaining a correspondence between the behavior state order similarity, the behavior state transition similarity, the behavior state value similarity, and the behavior sequence similarity;

[0024] According to the correspondence between the behavior state order similarity, the behavior state transition similarity, the behavior state value similarity, and the behavior sequence similarity, and a preset coefficient range, determine coefficient data;

[0025] According to the correspondence between the behavior state order similarity, the behavior state transition similarity, the behavior state value similarity, and the behavior sequence similarity, and the coefficient data, determine the behavior sequence similarity between the target user and the candidate user.

[0026] Optionally, the determining the user similarity between the target user and the candidate user according to the behavior sequence similarity comprises:

[0027] Obtaining the generation sequence time and behavior time interval of the target user and the generation sequence time and behavior time interval of the candidate user;

[0028] According to the generation sequence time of the target user, assign corresponding time weight data to the behavior sequence of the target user;

[0029] According to the generation sequence time and behavior time interval of the candidate user, assign corresponding time weight data to the candidate state sequence;

[0030] Obtaining a correspondence between the time weight data, the behavior sequence similarity, and the user similarity;

[0031] According to the correspondence between the time weight data, the behavior sequence similarity, and the user similarity, the time weight data of the target user, the time weight data of the candidate user, and the behavior sequence similarity, determine the user similarity between the target user and the candidate user.

[0032] Optionally, the obtaining the association weight data between the historical wrong questions of the target user and the candidate recommended test questions comprises:

[0033] obtaining an outer product of the historical wrong questions of the target user and the candidate recommended test questions;

[0034] splicing the outer product with the historical wrong questions and the candidate recommended test questions to obtain splicing data;

[0035] inputting the splicing data into a preset activation network composed of a preset activation function to obtain correlation weight data between the historical wrong questions and the candidate recommended test questions.

[0036] Optionally, the determining of the test question recommendation strategy according to the correlation weight data and the behavior sequence of the target user comprises:

[0037] obtaining a corresponding relationship between the correlation weight data and the behavior sequence of the target user;

[0038] determining the test question recommendation strategy according to the corresponding relationship between the correlation weight data and the behavior sequence of the target user, the correlation weight data and the behavior sequence of the target user.

[0039] In addition, to achieve the above object, the application further provides a test question recommendation device based on a behavior sequence, which comprises:

[0040] an obtaining module, configured to obtain a behavior sequence of a target user and a behavior sequence of a candidate user, and determine a behavior sequence similarity between the target user and the candidate user according to the behavior sequence of the target user and the behavior sequence of the candidate user;

[0041] the obtaining module is further configured to determine a user similarity between the target user and the candidate user according to the behavior sequence similarity;

[0042] a candidate module, configured to determine similar users of the target user in the candidate users according to a preset selection quantity and the user similarity, and obtain wrong questions of the similar users in a preset time range as candidate recommended test questions;

[0043] a recommendation module, configured to obtain correlation weight data between historical wrong questions of the target user and the candidate recommended test questions, and determine a test question recommendation strategy according to the correlation weight data and the behavior sequence of the target user;

[0044] the recommendation module is further configured to determine a target recommended test question in the candidate recommended test questions according to the test question recommendation strategy, and recommend the target recommended test question to the target user.

[0045] In addition, to achieve the above object, the application further provides a behavior sequence-based test question recommendation device, which comprises a memory, a processor and a behavior sequence-based test question recommendation program stored in the memory and executable on the processor, and the behavior sequence-based test question recommendation program is configured to implement the steps of the above behavior sequence-based test question recommendation method.

[0046] In addition, to achieve the above object, the application further provides a storage medium, which stores a behavior sequence-based test question recommendation program, and the behavior sequence-based test question recommendation program implements the steps of the above behavior sequence-based test question recommendation method when executed by a processor.

[0047] In the application, the behavior sequence of a target user and the behavior sequence of a candidate user are acquired, the behavior sequence similarity between the target user and the candidate user is determined according to the behavior sequence of the target user and the behavior sequence of the candidate user, the user similarity between the target user and the candidate user is determined according to the behavior sequence similarity, the similar users of the target user are determined in the candidate users according to the preset selection quantity and the user similarity, the wrong questions of the similar users in a preset time range are acquired as candidate recommended test questions, the association weight data between the historical wrong questions of the target user and the candidate recommended test questions are acquired, the test question recommendation strategy is determined according to the association weight data and the behavior sequence of the target user, the target recommended test question is determined in the candidate recommended test questions according to the test question recommendation strategy, and the target recommended test question is recommended to the target user. Compared with the fact that most online test question systems do not sufficiently dig the user portrait and the test question has strong hysteresis, the application can sufficiently dig the user features, find the similar users of the target user according to the historical behavior sequence of the target user, provide the candidate recommended test questions according to the historical wrong questions of the similar users, and use the attention mechanism to investigate the association between the candidate recommended test questions and the historical wrong questions of the target user, thereby improving the accuracy of the recommendation. In addition, the use of the recent historical behavior sequence of the target user can dynamically display the feature change of the target user, so that the user portrait is also dynamically updated, and obvious hysteresis does not occur, thereby realizing flexible personalized recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 FIG. 1 is a structural schematic diagram of a behavior sequence-based test question recommendation device of a hardware running environment related to an embodiment scheme of the application;

[0049] Figure 2 FIG. 2 is a flow schematic diagram of a first embodiment of a behavior sequence-based test question recommendation method of the application;

[0050] Figure 3 FIG. 3 is a whole flow schematic diagram of an embodiment of a behavior sequence-based test question recommendation method of the application;

[0051] Figure 4 Fig. 2 is a flowchart of a second embodiment of the behavior sequence-based test question recommendation method of the present application;

[0052] Figure 5 Fig. 3 is a flowchart of a third embodiment of the behavior sequence-based test question recommendation method of the present application;

[0053] Figure 6 Fig. 4 is a schematic diagram of an activation unit of the first embodiment of the behavior sequence-based test question recommendation method of the present application;

[0054] Figure 7 Fig. 5 is a structural block diagram of the first embodiment of the behavior sequence-based test question recommendation device of the present application.

[0055] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0056] It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

[0057] Reference Figure 1 , Figure 1 Fig. 6 is a structural schematic diagram of a behavior sequence-based test question recommendation device related to the hardware running environment of the embodiment of the present application.

[0058] As Figure 1 shown, the behavior sequence-based test question recommendation device can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display, an input unit such as a keyboard, and can also include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (Wi-Fi) interface). The memory 1005 can be a high-speed random access memory (RAM) memory, or a stable non-volatile memory (NVM), such as a disk memory. The memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0059] Those skilled in the art can understand, Figure 1The structure shown does not constitute a limitation on a test item recommendation device based on behavior sequence and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0060] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a test question recommendation program based on behavior sequences.

[0061] exist Figure 1 In the behavior sequence-based test item recommendation device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the behavior sequence-based test item recommendation device of the present invention can be set in the behavior sequence-based test item recommendation device, and the behavior sequence-based test item recommendation device calls the behavior sequence-based test item recommendation program stored in the memory 1005 through the processor 1001 and executes the behavior sequence-based test item recommendation method provided in the embodiment of the present invention.

[0062] This invention provides a test item recommendation method based on behavioral sequences, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of a test item recommendation method based on behavioral sequences according to the present invention.

[0063] In this embodiment, the test item recommendation method based on behavior sequence includes the following steps:

[0064] Step S10: Obtain the behavior sequence of the target user and the behavior sequence of the candidate user, and determine the behavior sequence similarity between the target user and the candidate user based on the behavior sequence of the target user and the behavior sequence of the candidate user.

[0065] It should be noted that the execution entity in this embodiment is an online question-solving system. The online question-solving system is equipped with a question recommendation program based on behavior sequence. The online question-solving system usually contains multiple functional modules, such as a question-solving module, a review of wrong questions module, and a statistics viewing module. The question recommendation program based on behavior sequence is used to recommend questions.

[0066] It is understood that the target user refers to the user for whom test questions need to be recommended, and the candidate user refers to other users besides the target user. This can be all other users or other users who may be similar users after preliminary screening. This embodiment does not limit this. Among them, similar users refer to users who are highly similar to the target user.

[0067] It should be understood that the behavior sequence includes several state string data arranged in a preset time order. The state string data refers to the user's state string, or user behavior, including access data and the operation data corresponding to the access data. The access data refers to the user's access to the functional modules of the online question-solving system, and the operation data refers to the user's operation in the functional modules. The i-th state string si is usually represented by (,yi), where zi represents the i-th functional module accessed by the user, and yi represents the corresponding operation when the user accesses the i-th functional module. The preset time order refers to the chronological order. The user's state strings are usually arranged in chronological order, resulting in a finite set {(z1,y1),(z2,y2)…(zn,yn)},≥2, which is the behavior sequence S. In this embodiment, the behavior sequence includes the target user's behavior sequence and the candidate user's behavior sequence, which can be represented by S respectively. i and S j This is represented by sim(S). Behavioral sequence similarity refers to the similarity of behavioral sequences between two users. In this embodiment, the behavioral sequence similarity is the similarity of behavioral sequences between the target user and the candidate user, used to determine the similarity between the target user and the candidate user. i ,S j ) is used to represent this.

[0068] In the specific implementation, the behavior sequence of the target user and the behavior sequence of the candidate user are first determined based on the user's access and operation of the functional modules. The similarity between the behavior sequences of the target user and the candidate user is then calculated.

[0069] Step S20: Determine the user similarity between the target user and the candidate user based on the behavioral sequence similarity.

[0070] Further, step S20 includes: obtaining the generation sequence time and behavior time interval of the target user and the generation sequence time and behavior time interval of the candidate user; assigning corresponding time weight data to the behavior sequence of the target user according to the generation sequence time of the target user; assigning corresponding time weight data to the candidate state sequence according to the generation sequence time and behavior time interval of the candidate user; obtaining the correspondence between the time weight data, the behavior sequence similarity and the user similarity; and determining the user similarity between the target user and the candidate user based on the correspondence between the time weight data, the behavior sequence similarity and the user similarity, the time weight data of the target user, the time weight data of the candidate user and the behavior sequence similarity.

[0071] It should be noted that the generation sequence time refers to the time from when the user first generates a behavior sequence to the currently generated user sequence, and the behavior time interval refers to the interval between each behavior in the behavior sequence. In this embodiment, the generation sequence time includes the generation sequence time of the target user and the generation sequence time of the candidate user. Correspondingly, the behavior time interval includes the behavior time interval of the target user and the behavior time interval of the candidate user.

[0072] It is understandable that user characteristics change dynamically over time. Therefore, when calculating the similarity between users based on behavioral sequence similarity, a time weighting coefficient needs to be introduced. More recent behavioral sequences are given a higher weighting coefficient, while behavioral sequences with longer time intervals have less reference value and are assigned a lower weighting coefficient. The time weighting data refers to the weighting coefficients assigned to behavioral sequences in this embodiment, including the time weighting data of the target user and the time weighting data of the candidate user. The time weighting data can be allocated according to the following calculation formula:

[0073]

[0074] In the formula, W(A,S) i This represents the time-weighted data for user A. L represents the time from when user A first generated a user sequence to when the current user sequence is generated. A S represents the time interval of user A's actions in this sequence of actions. i This represents the sequence of actions of user A, where 'a' is a real number between 0 and 1 (excluding 0 and 1), and needs to be determined according to L. A and The measurement is performed. In this embodiment, time-weighted data W(A,S) is assigned to target user A. i Assign time-weighted data W(B,S) to candidate user B. j ).

[0075] It should be understood that the user similarity refers to the similarity between two users; in this embodiment, the user similarity is the similarity between the target user and the candidate user. The correspondence between the time-weighted data, the behavioral sequence similarity, and the user similarity refers to the formula for calculating user similarity, as shown below:

[0076]

[0077] In the formula, sim(A,B) represents the user similarity between target user A and candidate user B, and W(A,S) represents the similarity between them. iW(B,S) represents the assignment of time-weighted data to target user A. j ) represents the allocation of time-weighted data to candidate user B, sim(S) i ,S j S represents the similarity of behavioral sequences between target user A and candidate user B. A With S B Let A and B represent the behavioral sequences of target user A and candidate user B, respectively. By substituting the obtained time-weighted data of the target user, the time-weighted data of the candidate user, and the behavioral sequence similarity into the above formula for calculating user similarity, the user similarity between the target user and the candidate user can be obtained.

[0078] In the specific implementation, firstly, time weight coefficients are introduced for the behavior sequences of the target user and the behavior sequences of the candidate user. Then, based on the assigned time weight coefficients and the similarity of the previously obtained behavior sequences, the user similarity between the target user and the candidate user is calculated.

[0079] Step S30: Based on the preset selection quantity and the user similarity, determine the similar users of the target user among the candidate users, and obtain the wrong questions of the similar users within a preset time range as candidate recommended test questions.

[0080] It should be noted that the "similar users" are candidate users with a high degree of similarity to the target user. The "preset selection number" refers to the number of similar users to be selected, which can typically be set to 5-10, or other values ​​can be set according to the actual situation. This embodiment does not impose any restrictions on this. The "preset time range" refers to a pre-set range for selecting test questions, usually a recent period, such as 10 days or one week. This can be adjusted according to actual needs, and this embodiment does not impose any restrictions on this. The "candidate recommended test questions" refer to the incorrect questions answered by similar users recently. These are candidate test questions with the potential to be recommended, and suitable test questions can be selected from them for subsequent recommendations.

[0081] In the specific implementation, similar users of the target user are identified, and the wrong answers of these similar users in the recent period are used as candidate recommended wrong answers to generate a candidate recommended test question set, which is used to determine the subsequent recommended test questions.

[0082] Step S40: Obtain the association weight data between the target user's historical incorrect questions and the candidate recommended questions, and determine the question recommendation strategy based on the association weight data and the target user's behavior sequence.

[0083] It is understood that the target user's historical incorrect questions refer to the target user's recent incorrect questions. The correlation weight data between the target user's historical incorrect questions and the candidate recommended questions refers to the degree of correlation between the historical incorrect questions and the candidate recommended questions. The greater the correlation, the more similar the candidate recommended questions are to the target user's incorrect questions, and the higher the probability that the target user will not answer the questions. The question recommendation strategy refers to the way questions are recommended. In this embodiment, the recommended questions are determined based on the degree of correlation between historical incorrect questions and candidate recommended questions, and usually, questions with a high degree of correlation are recommended to the target user.

[0084] It should be understood that an activation unit is provided in this embodiment. The activation unit has two inputs: one is a wrong question in the target user's historical behavior sequence, and the other is a candidate recommended question. It can output the correlation weight data between the target user's historical wrong questions and the candidate recommended questions.

[0085] Step S50: Based on the test question recommendation strategy, determine the target recommended test question from the candidate recommended test questions, and recommend the target recommended test question to the target user.

[0086] In a specific implementation, the target recommended test questions refer to the test questions that are ultimately recommended to the target user. In this embodiment, subsequent recommended test questions that are more closely related to the target user's historical wrong questions will be recommended to the target user to help the user quickly find weaknesses and fill in the gaps in their knowledge.

[0087] like Figure 3 The overall process diagram shown calculates the similarity of user behavior sequences based on the target user's behavior sequence to identify similar users. The recent incorrect questions of similar users are used as candidate recommended questions. A question that the target user answered incorrectly is used as input along with the candidate recommended questions to obtain a score, which is used to measure the correlation between the two. The higher the correlation between the selected question and the historical incorrect questions, the higher the probability that the user answered the question incorrectly, and thus the question can be recommended to the target user.

[0088] In this embodiment, the behavior sequences of the target user and candidate users are obtained. Based on these sequences, the similarity between the target user and the candidate users is determined. Then, the user similarity between the target user and the candidate users is determined based on the behavior sequence similarity. According to a preset selection number and the user similarity, similar users to the target user are identified from the candidate users. Incorrect answers from these similar users within a preset time range are obtained as candidate recommended questions. The association weight data between the target user's historical incorrect answers and the candidate recommended questions is obtained. Based on the association weight data and the target user's behavior sequence, a question recommendation strategy is determined. According to the question recommendation strategy, a target recommended question is identified from the candidate recommended questions and recommended to the target user. This embodiment can fully leverage user characteristics, find similar users based on the user's historical behavior sequences, provide candidate recommended questions based on the historical incorrect answers of similar users, and utilize an attention mechanism to examine the correlation between candidate recommended questions and the user's historical incorrect answers, improving the accuracy of recommendations. Furthermore, using the user's recent historical behavior sequences dynamically displays changes in user characteristics, ensuring that the user profile is dynamically updated without significant lag, achieving flexible personalized recommendations.

[0089] Reference Figure 4 , Figure 4 This is a flowchart illustrating a second embodiment of a test item recommendation method based on behavioral sequences according to the present invention.

[0090] Based on the above embodiments, step S10 includes:

[0091] Step S101: Concatenate the state string data in the behavior sequence of the target user to obtain the behavior state sequence of the target user; concatenate the state string data in the behavior sequence of the candidate user to obtain the behavior state sequence of the candidate user.

[0092] It should be noted that the behavior state sequence is a string formed by concatenating all state strings in order based on the behavior sequence. For example, when the behavior sequence is {s1,s2,…,sn}, and ≥2, the behavior state sequence is s1s2…sn.

[0093] In the specific implementation, the state strings in the behavior sequences of the target user and the candidate user are concatenated sequentially to form a behavior-state sequence B. i and B j .

[0094] Step S102: Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the similarity of behavioral state order, behavioral state transition, and behavioral state value between the target user and the candidate user.

[0095] Further, step S102 includes:

[0096] Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the maximum common state subsequence and determine the length of the maximum common state subsequence corresponding to the maximum common state subsequence.

[0097] It is understandable that the similarity between two behavioral state sequences in the order of occurrence is called the state order similarity. In order to find the state order similarity, it is necessary to first find their maximum common state subsequence, that is, the maximum common state subsequence between the behavioral state sequence of the target user and the behavioral state sequence of the candidate user. The length of the maximum common state subsequence refers to the sequence length of the maximum common state subsequence.

[0098] Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the number of common state transitions, the number of state transitions, the number of state strings, and the number of common state strings.

[0099] It should be understood that the process of a user's state string changing is called a state transition. For example, the transition from si to sj is a user state transition. A user's state transition can indirectly reflect the trend of the user's attention shift. By judging and analyzing similar state transitions, we can also determine the similarity between users.

[0100] It should be noted that the common state transition count refers to the sum of the number of times that two behavioral state sequences have the same state transitions, the state transition sum count refers to the sum of the number of state transitions in the two behavioral state sequences, the common state string count refers to the number of common state strings contained in the two behavioral state sequences, and the state string sum count refers to the sum of the number of state strings in the two behavioral state sequences.

[0101] The similarity of the behavior state order is determined based on the correspondence between the length of the maximum common state subsequence, the number of state strings and the similarity of the behavior state order, the length of the maximum common state subsequence, and the number of state strings.

[0102] It is understood that the correspondence between the length of the maximum common state subsequence, the number of state strings, and the similarity of the behavior state order refers to the formula for calculating the similarity of the state order, as shown below:

[0103]

[0104] In the formula, sim_seq(B i B j B represents the sequence of behavioral states of the target user. iWith the candidate user's behavioral state sequence B j The similarity of the state order between them, len[comm(B i B j )] represents B i With B j The length of the longest common state subsequence between them, comm(B) i B j ) represents B i With B j The longest common subsequence of states between |B i ∪B j | indicates B i With B j The length and number of the maximum common state subsequence are used to calculate the behavioral state order similarity. By substituting the length of the maximum common state subsequence, the number of the state strings, and the number of states into the above formula for calculating state order similarity, the behavioral state order similarity can be calculated.

[0105] The similarity of the behavioral state transitions is determined based on the number of common state transitions, the correspondence between the number of state transitions and the similarity of the behavioral state transitions, the number of common state transitions, and the number of state transitions.

[0106] It should be understood that the state transition similarity refers to the similarity of state transitions in two behavioral state sequences, and the correspondence between the number of common state transitions, the number of state transitions, and the behavioral state transition similarity refers to the formula for calculating the state transition similarity, as shown below:

[0107]

[0108] In the formula, sim_trans(B i B j B represents the sequence of behavioral states of the target user. i With the candidate user's behavioral state sequence B j The state transition similarity between them, ω+δ represents B i With B j The state transitions and quantities between them B i With B j The number of common state transitions between the states. Substituting the number of common state transitions and the sum of state transitions into the above formula for calculating state transition similarity, we can obtain the state transition similarity.

[0109] The similarity of the behavior state value is determined based on the correspondence between the state string and its quantity, the number of common state strings and the similarity of the behavior state value, the state string and its quantity, and the number of common state strings.

[0110] It should be noted that the behavioral state value similarity refers to the similarity of state values ​​in two behavioral state sequences. The correspondence between the number of state strings, the number of common state strings, and the behavioral state value similarity refers to the calculation formula for behavioral state value similarity, as shown below:

[0111]

[0112] In the formula, sim_value(B i B j B represents the sequence of behavioral states of the target user. i With the candidate user's behavioral state sequence B j The similarity of behavioral state values ​​between them, |B i ∪B j | indicates B i With B j The state string and its count, |B i ∩B j | indicates B i With B j The number of common state strings between them. Substituting the number of state strings and their corresponding number of common state strings into the above formula for calculating the similarity of behavioral state values, we can obtain the similarity of behavioral state values.

[0113] In the specific implementation, the maximum common state subsequence, the number of common state transitions, the number of state transitions, the number of state strings, and the number of common state strings are first determined, so as to calculate the similarity of the behavioral state order, the similarity of behavioral state transitions, and the similarity of behavioral state values ​​between the target user and the candidate user.

[0114] Step S103: Determine the behavioral sequence similarity between the target user and the candidate user based on the behavioral state order similarity, the behavioral state transition similarity, and the behavioral state value similarity.

[0115] Further, step S103 includes: obtaining the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity; determining coefficient data based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity, and a preset coefficient range; and determining the behavioral sequence similarity between the target user and the candidate user based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity, and the coefficient data.

[0116] It is understood that the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity refers to the calculation formula for behavioral sequence similarity, as shown below:

[0117] sim(B i B j )=α×sim_seq(B i B j )+β×sim_trans(B i B j )+γ×sim_value(B i B j )

[0118] In the formula, sim(B i B j B represents the sequence of behavioral states of the target user. i With the candidate user's behavioral state sequence B j The similarity of behavioral sequences between them, sim_seq(B i B j ) represents B i With B j The similarity of the state order between them, im_trans(B i B j ) represents B i With B j The state transition similarity between them, sim_value(B) i B j ) represents B i With B j The similarity of behavioral state values ​​between users is calculated using coefficients α, β, and γ. The preset coefficient range refers to the constraints on the values ​​of α, β, and γ. In this embodiment, the sum of α, β, and γ is always equal to 1, and all are greater than or equal to 0. These values ​​are continuously adjusted during the calculation of behavioral sequence similarity to obtain optimal values. The calculated values ​​are proportional to the similarity between users. Substituting the obtained coefficient data, behavioral state order similarity, behavioral state transition similarity, and behavioral state value similarity into the above-mentioned formula for calculating behavioral sequence similarity yields the behavioral sequence similarity.

[0119] In this embodiment, the target user's behavior state sequence is obtained by sequentially concatenating the state string data in the target user's behavior sequence, and the candidate user's behavior state sequence is obtained by sequentially concatenating the state string data in the candidate user's behavior sequence. Based on the target user's behavior state sequence and the candidate user's behavior state sequence, the similarity of behavior state order, behavior state transition, and behavior state value between the target user and the candidate user is determined. Based on the similarity of behavior state order, behavior state transition, and behavior state value, the similarity of behavior sequence between the target user and the candidate user is determined. This embodiment can fully explore user characteristics and find similar users based on the user's historical behavior sequence, improving the accuracy of recommendations. In addition, using the user's recent historical behavior sequence can dynamically display changes in user characteristics, making the user profile dynamically updated without significant lag, and realizing flexible personalized recommendations.

[0120] Reference Figure 5 , Figure 5 This is a flowchart illustrating a third embodiment of a test item recommendation method based on behavioral sequences according to the present invention.

[0121] Based on the above embodiments, step S40 includes:

[0122] Step S401: Obtain the outer product of the target user's historical incorrect questions and the candidate recommended questions, and concatenate the outer product with the historical incorrect questions and the candidate recommended questions to obtain concatenated data.

[0123] It should be noted that the spliced ​​data refers to the data obtained by splicing the outer product with historical wrong questions and candidate recommended questions.

[0124] It is understood that the historical incorrect questions and candidate recommended questions in this embodiment can be considered as the relevant data corresponding to the historical incorrect questions and candidate recommended questions.

[0125] Step S402: Input the spliced ​​data into a preset activation network to obtain the association weight data between the historical wrong questions and the candidate recommended questions.

[0126] It should be understood that the preset activation network refers to the multi-layer network used in the activation unit. The preset activation network is composed of preset activation functions, which are pre-set activation functions. In this embodiment, the preset activation functions used are PRelu (Parametric Rectified Linear Unit) / Dice (Dice Activation Function) and Linear (linear activation function). They can also be adjusted according to actual needs. This embodiment does not limit this.

[0127] Step S403: Obtain the correspondence between the associated weight data and the target user's behavior sequence, and determine the test question recommendation strategy based on the correspondence between the associated weight data and the target user's behavior sequence, the associated weight data, and the target user's behavior sequence.

[0128] It should be noted that the correspondence between the associated weight data and the target user's behavior sequence refers to the expression of the candidate recommended questions, which can be obtained by multiplying the obtained associated weight data with the historical incorrect questions, as shown below:

[0129]

[0130] In the formula, v U (A) represents the expression of target user U for candidate recommended question A, {e1,e2,…,e H} represents the sequence of actions of target user U, a(e j ,v A ) represents the association weight corresponding to the i-th historical wrong question.

[0131] In the specific implementation, by substituting the associated weight data and the target user's behavior sequence into the expression of the above candidate recommended questions, the expression of the target user for the candidate recommended questions can be obtained, and the question recommendation strategy can be determined.

[0132] like Figure 6 As shown, the activation unit has two inputs: one is a wrong question in the target user's behavior sequence, and the other is a candidate recommendation question. The two are concatenated with their outer product and then input into a multi-layer network to learn a weight corresponding to the historical wrong question. Finally, the obtained weight is multiplied with the historical wrong question to obtain the target user's expression for the candidate recommendation question.

[0133] In the specific implementation, the outer product of the target user's historical incorrect questions and the candidate recommended questions is obtained. This outer product is then concatenated with the historical incorrect questions and the candidate recommended questions to obtain concatenated data. This concatenated data is input into a preset activation network to obtain the association weight data between the historical incorrect questions and the candidate recommended questions. The correspondence between the association weight data and the target user's behavior sequence is then obtained. Based on the correspondence between the association weight data and the target user's behavior sequence, the association weight data, and the target user's behavior sequence, a question recommendation strategy is determined. This embodiment utilizes an attention mechanism to examine the correlation between candidate recommended questions and the user's historical incorrect questions. If the correlation is strong, it indicates that the candidate recommended questions are very similar to the target user's historical incorrect questions, and the user is highly likely not to answer the questions correctly. This helps the user quickly identify weaknesses and fill in knowledge gaps, improving the accuracy of the recommendations.

[0134] Furthermore, this embodiment of the invention also proposes a storage medium storing a test item recommendation program based on behavior sequences. When the test item recommendation program based on behavior sequences is executed by a processor, it implements the steps of the test item recommendation method based on behavior sequences as described above.

[0135] Reference Figure 7 , Figure 7 This is a structural block diagram of the first embodiment of the test item recommendation device based on behavior sequence of the present invention.

[0136] like Figure 7 As shown, the test item recommendation device based on behavior sequence proposed in this embodiment of the invention includes:

[0137] The acquisition module 10 is used to acquire the behavior sequence of the target user and the behavior sequence of the candidate user, and determine the behavior sequence similarity between the target user and the candidate user based on the behavior sequence of the target user and the behavior sequence of the candidate user.

[0138] The acquisition module 10 is further configured to determine the user similarity between the target user and the candidate user based on the behavioral sequence similarity.

[0139] The candidate module 20 is used to determine similar users of the target user from the candidate users based on the preset selection quantity and the similarity of the user, and to obtain the wrong questions of the similar users within a preset time range as candidate recommended test questions.

[0140] The recommendation module 30 is used to obtain the association weight data between the target user's historical wrong questions and the candidate recommended questions, and to determine the question recommendation strategy based on the association weight data and the target user's behavior sequence.

[0141] The recommendation module 30 is further configured to determine a target recommended question from the candidate recommended questions according to the question recommendation strategy, and recommend the target recommended question to the target user.

[0142] In this embodiment, the behavior sequences of the target user and candidate users are obtained. Based on these sequences, the similarity between the target user and the candidate users is determined. Then, the user similarity between the target user and the candidate users is determined based on the behavior sequence similarity. According to a preset selection number and the user similarity, similar users to the target user are identified from the candidate users. Incorrect answers from these similar users within a preset time range are obtained as candidate recommended questions. The association weight data between the target user's historical incorrect answers and the candidate recommended questions is obtained. Based on the association weight data and the target user's behavior sequence, a question recommendation strategy is determined. According to the question recommendation strategy, a target recommended question is identified from the candidate recommended questions and recommended to the target user. This embodiment can fully leverage user characteristics, find similar users based on the user's historical behavior sequences, provide candidate recommended questions based on the historical incorrect answers of similar users, and utilize an attention mechanism to examine the correlation between candidate recommended questions and the user's historical incorrect answers, improving the accuracy of recommendations. Furthermore, using the user's recent historical behavior sequences dynamically displays changes in user characteristics, ensuring that the user profile is dynamically updated without significant lag, achieving flexible personalized recommendations.

[0143] In one embodiment, the behavior sequence includes several state string data arranged in a preset time order. The state string data includes access data and operation data corresponding to the access data. The acquisition module 10 is also used to connect the state string data in the behavior sequence of the target user in sequence to obtain the behavior state sequence of the target user.

[0144] The state string data in the behavior sequence of the candidate user are concatenated sequentially to obtain the behavior state sequence of the candidate user;

[0145] Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the similarity of behavioral state order, behavioral state transition, and behavioral state value between the target user and the candidate user.

[0146] The behavioral sequence similarity between the target user and the candidate user is determined based on the behavioral state order similarity, the behavioral state transition similarity, and the behavioral state value similarity.

[0147] In one embodiment, the acquisition module 10 is further configured to determine the maximum common state subsequence based on the behavior state sequence of the target user and the behavior state sequence of the candidate user, and to determine the length of the maximum common state subsequence corresponding to the maximum common state subsequence;

[0148] Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the number of common state transitions, the number of state transitions, the number of state strings, and the number of common state strings.

[0149] The behavioral state order similarity is determined based on the correspondence between the length of the maximum common state subsequence, the number of state strings and the similarity of the behavioral state order, the length of the maximum common state subsequence, and the number of state strings.

[0150] The behavioral state transition similarity is determined based on the number of common state transitions, the correspondence between the number of state transitions and the behavioral state transition similarity, the number of common state transitions, and the number of state transitions.

[0151] The similarity of the behavior state value is determined based on the correspondence between the state string and its quantity, the number of common state strings and the similarity of the behavior state value, the state string and its quantity, and the number of common state strings.

[0152] In one embodiment, the acquisition module 10 is further configured to acquire the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity;

[0153] Based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity and the behavioral sequence similarity, and the preset coefficient range, the coefficient data is determined;

[0154] The behavioral sequence similarity between the target user and the candidate user is determined based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity and the behavioral sequence similarity, and the coefficient data.

[0155] In one embodiment, the acquisition module 10 is further configured to acquire the generation sequence time and behavior time interval of the target user and the generation sequence time and behavior time interval of the candidate user;

[0156] Based on the generation sequence time of the target user, assign corresponding time weight data to the behavioral sequence of the target user;

[0157] Based on the generation sequence time and behavior time interval of the candidate users, the candidate state sequence is assigned corresponding time weight data;

[0158] Obtain the correspondence between the time weight data, the behavior sequence similarity, and the user similarity;

[0159] The user similarity between the target user and the candidate user is determined based on the time-weighted data, the correspondence between the behavioral sequence similarity and the user similarity, the time-weighted data of the target user, the time-weighted data of the candidate user, and the behavioral sequence similarity.

[0160] In one embodiment, the recommendation module 30 is further configured to obtain the cross product of the target user's historical incorrect questions and the candidate recommended questions;

[0161] The outer product is concatenated with the historical incorrect questions and the candidate recommended questions to obtain concatenated data;

[0162] The spliced ​​data is input into a preset activation network to obtain the association weight data between the historical wrong questions and the candidate recommended questions. The preset activation network is composed of preset activation functions.

[0163] In one embodiment, the recommendation module 30 is further configured to obtain the correspondence between the associated weight data and the behavioral sequence of the target user;

[0164] The test question recommendation strategy is determined based on the correspondence between the associated weight data and the target user's behavior sequence, the associated weight data, and the target user's behavior sequence.

[0165] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0166] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0167] In addition, for technical details not described in detail in this embodiment, please refer to the test item recommendation method based on behavior sequence provided in any embodiment of the present invention, which will not be repeated here.

[0168] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0169] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0171] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A test item recommendation method based on behavior sequences, characterized in that, The test item recommendation method based on behavior sequences includes: Obtain the behavior sequence of the target user and the behavior sequence of the candidate user, and determine the behavior sequence similarity between the target user and the candidate user based on the behavior sequence of the target user and the behavior sequence of the candidate user; Based on the behavioral sequence similarity, the user similarity between the target user and the candidate user is determined; Based on the preset selection quantity and the user similarity, similar users to the target user are determined from the candidate users, and the wrong questions of the similar users within a preset time range are obtained as candidate recommended test questions; Obtain the association weight data between the target user's historical incorrect questions and the candidate recommended questions, and determine the question recommendation strategy based on the association weight data and the target user's behavior sequence; According to the test question recommendation strategy, a target test question is determined from the candidate recommended test questions, and the target recommended test question is recommended to the target user; Determining the user similarity between the target user and the candidate user based on the behavioral sequence similarity includes: Obtain the generation sequence time and behavior time interval of the target user, as well as the generation sequence time and behavior time interval of the candidate user; Based on the generation sequence time of the target user, assign corresponding time weight data to the behavioral sequence of the target user; Based on the generation sequence time and behavior time interval of the candidate users, the candidate state sequence is assigned corresponding time weight data; Obtain the correspondence between the time weight data, the behavior sequence similarity, and the user similarity; Based on the time-weighted data, the correspondence between the behavioral sequence similarity and the user similarity, the time-weighted data of the target user, the time-weighted data of the candidate user, and the behavioral sequence similarity, the user similarity between the target user and the candidate user is determined. The step of obtaining the association weight data between the target user's historical incorrect questions and the candidate recommended questions includes: Obtain the outer product of the target user's historical incorrect answers and the candidate recommended questions; The outer product is concatenated with the historical incorrect questions and the candidate recommended questions to obtain concatenated data; The spliced ​​data is input into a preset activation network to obtain the association weight data between the historical wrong questions and the candidate recommended questions. The preset activation network is composed of preset activation functions.

2. The method as described in claim 1, characterized in that, The behavior sequence includes several state string data arranged in a preset time order. The state string data includes access data and operation data corresponding to the access data. Determining the similarity of the behavior sequences between the target user and the candidate user based on the behavior sequences of the target user and the candidate user includes: The state string data in the target user's behavior sequence are concatenated sequentially to obtain the target user's behavior state sequence. The state string data in the behavior sequence of the candidate user are concatenated sequentially to obtain the behavior state sequence of the candidate user; Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the similarity of behavioral state order, behavioral state transition, and behavioral state value between the target user and the candidate user. The behavioral sequence similarity between the target user and the candidate user is determined based on the behavioral state order similarity, the behavioral state transition similarity, and the behavioral state value similarity.

3. The method as described in claim 2, characterized in that, The step of determining the similarity of behavioral state order, behavioral state transition, and behavioral state value between the target user and the candidate user based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user includes: Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the maximum common state subsequence and determine the length of the maximum common state subsequence corresponding to the maximum common state subsequence; Based on the behavioral state sequence of the target user and the behavioral state sequence of the candidate user, determine the number of common state transitions, the number of state transitions, the number of state strings, and the number of common state strings. The behavioral state order similarity is determined based on the correspondence between the length of the maximum common state subsequence, the number of state strings and the similarity of the behavioral state order, the length of the maximum common state subsequence, and the number of state strings. The behavioral state transition similarity is determined based on the number of common state transitions, the correspondence between the number of state transitions and the behavioral state transition similarity, the number of common state transitions, and the number of state transitions. The similarity of the behavior state value is determined based on the correspondence between the state string and its quantity, the number of common state strings and the similarity of the behavior state value, the state string and its quantity, and the number of common state strings.

4. The method as described in claim 2, characterized in that, Determining the behavioral sequence similarity between the target user and the candidate user based on the behavioral state order similarity, the behavioral state transition similarity, and the behavioral state value similarity includes: Obtain the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity, and the behavioral sequence similarity; Based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity and the behavioral sequence similarity, and the preset coefficient range, the coefficient data is determined; The behavioral sequence similarity between the target user and the candidate user is determined based on the correspondence between the behavioral state order similarity, the behavioral state transition similarity, the behavioral state value similarity and the behavioral sequence similarity, and the coefficient data.

5. The method as described in claim 1, characterized in that, The step of determining the test question recommendation strategy based on the association weight data and the target user's behavior sequence includes: Obtain the correspondence between the associated weight data and the target user's behavior sequence; The test question recommendation strategy is determined based on the correspondence between the associated weight data and the target user's behavior sequence, the associated weight data, and the target user's behavior sequence.

6. A test item recommendation device based on behavior sequence, characterized in that, The test item recommendation device based on behavior sequence includes: The acquisition module is used to acquire the behavior sequence of the target user and the behavior sequence of the candidate user, and determine the behavior sequence similarity between the target user and the candidate user based on the behavior sequence of the target user and the behavior sequence of the candidate user; The acquisition module is further configured to determine the user similarity between the target user and the candidate user based on the behavioral sequence similarity; The candidate module is used to determine similar users of the target user from the candidate users based on the preset number of selections and the similarity of the user, and to obtain the wrong questions of the similar users within a preset time range as candidate recommended test questions; The recommendation module is used to obtain the association weight data between the target user's historical wrong questions and the candidate recommended questions, and to determine the question recommendation strategy based on the association weight data and the target user's behavior sequence. The recommendation module is further configured to determine a target recommended question from the candidate recommended questions according to the question recommendation strategy, and recommend the target recommended question to the target user; The acquisition module is further configured to acquire the generation sequence time and behavior time interval of the target user and the generation sequence time and behavior time interval of the candidate user; Based on the generation sequence time of the target user, assign corresponding time weight data to the behavioral sequence of the target user; Based on the generation sequence time and behavior time interval of the candidate users, the candidate state sequence is assigned corresponding time weight data; Obtain the correspondence between the time weight data, the behavior sequence similarity, and the user similarity; Based on the time-weighted data, the correspondence between the behavioral sequence similarity and the user similarity, the time-weighted data of the target user, the time-weighted data of the candidate user, and the behavioral sequence similarity, the user similarity between the target user and the candidate user is determined. The recommendation module is also used to obtain the cross product of the target user's historical wrong questions and the candidate recommended questions; The outer product is concatenated with the historical incorrect questions and the candidate recommended questions to obtain concatenated data; The spliced ​​data is input into a preset activation network to obtain the association weight data between the historical wrong questions and the candidate recommended questions. The preset activation network is composed of preset activation functions.

7. A test item recommendation device based on behavioral sequences, characterized in that, The device includes: a memory, a processor, and a behavior sequence-based test item recommendation program stored in the memory and executable on the processor, the behavior sequence-based test item recommendation program being configured to implement the steps of the behavior sequence-based test item recommendation method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium stores a test item recommendation program based on behavior sequences, which, when executed by a processor, implements the steps of the test item recommendation method based on behavior sequences as described in any one of claims 1 to 5.

Images