Path planning method and system based on user cultural tourism demand analysis

Abstract

This invention provides a path planning method and system based on user cultural and tourism demand analysis, belonging to the field of data processing technology. In this invention, firstly, user cultural and tourism demand information provided by the target user is obtained; secondly, at least one demand representation information corresponding to at least one demand sub-information included in the user cultural and tourism demand information is determined; then, a random semantic vector is determined; further, the random semantic vector is semantically encoded at multiple time steps, wherein, in the semantic encoding process of each time step, the semantic information carried by at least one demand representation information is fused into the random semantic vector; finally, the demand-encoded semantic vector formed by the semantic encoding of the last time step is semantically decoded to form the target planned path corresponding to the user cultural and tourism demand information. Based on the above method, the problem of relatively low reliability in path planning for cultural and tourism demand in existing technologies can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a path planning method and system based on user cultural and tourism demand analysis. Background Technology

[0002] With the rapid development of the tourism industry, tourists' demands for travel experiences are becoming increasingly personalized and diversified. Cultural and tourism demand path planning, as a crucial link in tourism information acquisition, itinerary arrangement, and experience optimization, directly impacts tourist satisfaction and experience outcomes. However, current technologies typically extract keywords from the tourist's input demand information, then perform text matching based on these keywords to identify corresponding cultural and tourism nodes, and finally generate the appropriate cultural and tourism path. This approach has two drawbacks. First, relying solely on keyword matching makes it difficult to fully utilize the semantic information of the tourist's input demand information, resulting in relatively low reliability of the path planning. Second, the inherent limitations of the tourist's input demand information also make it difficult to guarantee the reliability of the path planning. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a route planning method and system based on user cultural and tourism demand analysis, so as to improve the problem of relatively low reliability of route planning for cultural and tourism demand in the prior art.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0005] A route planning method based on user cultural tourism demand analysis includes:

[0006] Obtain information on the cultural and tourism needs of target users;

[0007] Determine at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information, wherein each demand representation information includes a demand sub-information and an emotion-related information corresponding to the demand sub-information, and the emotion-related information is obtained by analyzing the provision operations performed by the target user for the demand sub-information;

[0008] A random semantic vector is determined, wherein the random semantic vector is either randomly generated or a semantic vector formed by integrating other cultural and tourism related information on the basis of a randomly generated semantic vector;

[0009] The random semantic vector is semantically encoded at multiple time steps, wherein, in the semantic encoding process at each time step, the semantic information carried by at least one demand representation information is fused into the random semantic vector;

[0010] The semantic vector of demand encoding formed by the semantic encoding of the last time step is semantically decoded to form the target planning path corresponding to the user's cultural and tourism demand information.

[0011] In some preferred embodiments, in the above-described path planning method based on user cultural and tourism demand analysis, the step of determining at least one demand representation information corresponding to at least one demand sub-information included in the user cultural and tourism demand information includes:

[0012] Based on the similarity between text content, the user's cultural and tourism demand information is segmented to form at least one demand sub-information.

[0013] For each of the aforementioned demand sub-information, the input rate of the demand sub-information is determined based on the number of characters or words in the demand sub-information and the input duration of the demand sub-information; the input stability of the demand sub-information is determined based on the number of characters or words in the demand sub-information and the number of characters or words deleted during the input process; and the distribution dispersion is determined based on the input interval duration between each pair of adjacent characters or words in the demand sub-information.

[0014] The input rate, input stability, and distribution dispersion of the demand sub-information are determined as the emotion-related information corresponding to the demand sub-information. Based on the demand sub-information and the emotion-related information corresponding to the demand sub-information, a corresponding demand representation information is constructed.

[0015] In some preferred embodiments, in the above-described path planning method based on user cultural and tourism demand analysis, the step of determining the random semantic vector includes:

[0016] Randomly generate an initial semantic vector;

[0017] The cultural and tourism corpus is mapped into a vector space to form a cultural and tourism corpus mapping vector. The cultural and tourism corpus includes multiple cultural and tourism projects and the project content of each cultural and tourism project. Each path planning node in the target planning path belongs to one of the multiple cultural and tourism projects.

[0018] The initial semantic vector and the cultural tourism corpus mapping vector are matrix multiplied to form a relevance parameter distribution. Based on the target masking matrix, the relevance parameter distribution is locally masked to form a masking parameter distribution. The target masking matrix is ​​formed during the learning process of the corresponding neural network model through sample cultural tourism demand information and planned path labels.

[0019] The masking parameter distribution is filtered by the maximum value of the column and row or the mean value of the column and row to form a first filtering distribution and a second filtering distribution.

[0020] The first screening distribution and the second screening distribution are combined to form a random semantic vector.

[0021] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of semantically encoding the random semantic vector at multiple time steps includes:

[0022] Based on at least one demand sub-information, the first local vector in the random semantic vector is semantically encoded to form at least one demand encoding vector of the first local vector.

[0023] Based on at least one demand encoding vector and at least one emotion-related information of the first local vector, the emotion encoding vector of the first local vector is determined.

[0024] Based on the at least one demand sub-information, the second local vector in the random semantic vector is semantically encoded to form at least one demand encoding vector of the second local vector, wherein the first local vector and the second local vector partially overlap, and both the first local vector and the second local vector belong to a part of the random semantic vector;

[0025] Based on at least one demand encoding vector of the second local vector and the at least one emotion-related information, the emotion encoding vector of the second local vector is determined.

[0026] Based on the emotion encoding vector of the first local vector and the part of the emotion encoding vector of the second local vector that does not overlap with the first local vector, the demand encoding semantic vector of the last time step is formed.

[0027] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of semantically encoding a first local vector in the random semantic vector based on at least one demand sub-information to form at least one demand-encoded vector of the first local vector includes:

[0028] Perform vector space mapping on any one of the demand sub-informations to form a demand mapping vector corresponding to that one demand sub-information;

[0029] In the first fusion stage, based on the first local vector in the random semantic vector, cross-attention processing is performed on the demand mapping vector corresponding to any demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the first fusion stage.

[0030] In the second fusion stage, based on the fusion output vector of the first fusion stage, cross-attention processing is performed on the demand mapping vector corresponding to any one of the demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the second fusion stage.

[0031] The fusion output vector of the last fusion stage is determined as a requirement encoding vector of the first local vector.

[0032] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of determining the emotion encoding vector of the first local vector based on at least one demand encoding vector and at least one emotion-related information vector of the first local vector includes:

[0033] Each emotion-related information is mapped into a vector space to form an emotion mapping vector corresponding to each emotion-related information.

[0034] In at least one demand encoding vector of the first local vector, based on the correspondence between the demand sub-information and emotion-related information included in the same demand representation information, the demand encoding vector corresponding to each of the emotion mapping vectors is determined respectively.

[0035] For each of the aforementioned emotion mapping vectors, the emotion mapping vector and the corresponding demand encoding vector are interactively gated to form a first gated fusion vector and a second gated fusion vector.

[0036] The first gated fusion vector, the second gated fusion vector, the emotion mapping vector, and the demand encoding vector are connected to form the corresponding connection result;

[0037] When there are multiple emotion mapping vectors, the connection results corresponding to the multiple emotion mapping vectors are added together to form the emotion encoding vector of the first local vector.

[0038] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of interactively gating the emotion mapping vector and the corresponding demand encoding vector to form a first gating fusion vector and a second gating fusion vector for each emotion mapping vector includes:

[0039] The emotion mapping vector is subjected to linear mapping and nonlinear processing to form the first gating parameter;

[0040] The demand encoding vector corresponding to the emotion mapping vector is linearly mapped and nonlinearly processed to form the second gating parameter;

[0041] Multiply the parameters at corresponding positions between the first gating parameter and the demand encoding vector corresponding to the emotion mapping vector to form the first gating fusion vector;

[0042] The second gating parameter and the corresponding parameters between the emotion mapping vector are multiplied to form the second gating fusion vector.

[0043] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of performing linear mapping and nonlinear processing on the emotion mapping vector to form a first gating parameter includes:

[0044] The emotion mapping vector and its transpose are multiplied to form an emotion attention distribution. Based on the emotion attention distribution, the emotion mapping vector is weighted and summed to form an emotion attention vector.

[0045] Based on the target weight matrix and the target bias parameter, the emotion attention vector is linearly mapped to form an emotion linear vector; and based on the target nonlinear function, the emotion linear vector is nonlinearly mapped to form an emotion nonlinear vector.

[0046] The emotion nonlinear vector is determined as the first gating parameter.

[0047] In some preferred embodiments, in the above-described path planning method based on user cultural tourism demand analysis, the step of semantically encoding the random semantic vector at multiple time steps further includes:

[0048] The target segmentation bounding box, target movement direction, target segmentation scale, and original vector coordinates for local vector segmentation in the random semantic vector are determined.

[0049] When the target segmentation box is located at the original vector coordinates, the local vector currently enclosed by the target segmentation box in the random semantic vector is determined as the first local vector;

[0050] In the random semantic vector, starting from the original vector coordinates, the target segmentation box is moved a distance represented by the target segmentation scale along the target movement direction. Then, the local vector currently enclosed by the target segmentation box in the random semantic vector is determined as the second local vector.

[0051] This invention also provides a route planning system based on user cultural tourism demand analysis, comprising:

[0052] A processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described path planning method based on user cultural and tourism demand analysis.

[0053] The path planning method and system based on user cultural tourism demand analysis provided by this invention first obtains user cultural tourism demand information provided by the target user; second, it determines at least one demand representation information corresponding to at least one demand sub-information included in the user cultural tourism demand information; then, it determines a random semantic vector; further, it performs semantic encoding on the random semantic vector at multiple time steps, wherein, in the semantic encoding process of each time step, the semantic information carried by at least one demand representation information is fused into the random semantic vector; finally, it performs semantic decoding on the demand-encoded semantic vector formed by the semantic encoding of the last time step to form the target planning path corresponding to the user cultural tourism demand information. Based on the above method, firstly, because semantic encoding is performed, the potential semantic information in the user cultural tourism demand information can be mined, thereby obtaining a target planning path that is more matched to the actual cultural tourism demand of the target user. Secondly, because random semantic vectors are used in the semantic encoding process, overfitting can be avoided in the semantic encoding process, and the user cultural tourism demand information and potential possibilities provided by the target user can be better balanced in the semantic decoding process, thereby generating a more accurate and high-quality path. Thirdly, because the semantic encoding process not only mines the latent semantic information of the user's cultural and tourism needs themselves, but also mines the latent semantic information related to the emotions reflected in the target user's actions, the resulting demand-encoded semantic vector can represent the target user's most authentic needs. Therefore, it can improve the problem of relatively low reliability in the path planning of cultural and tourism needs in existing technologies.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0055] Figure 1 This is a structural block diagram of a route planning system based on user cultural and tourism demand analysis provided in an embodiment of the present invention.

[0056] Figure 2 This is a flowchart illustrating the steps of the path planning method based on user cultural and tourism demand analysis provided in this embodiment of the invention.

[0057] Figure 3 This is a schematic diagram illustrating the determination of a random semantic vector according to an embodiment of the present invention.

[0058] Figure 4This is a schematic diagram illustrating the semantic encoding of multiple time steps provided in an embodiment of the present invention.

[0059] Figure 5 This is a schematic diagram of the semantic encoding of the first local vector provided in an embodiment of the present invention.

[0060] Figure 6 This is a schematic diagram illustrating the fusion of emotion-related information provided in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0062] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0063] like Figure 1 As shown, this embodiment of the invention provides a route planning system based on user cultural and tourism demand analysis. The route planning system based on user cultural and tourism demand analysis may include a memory and a processor.

[0064] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the path planning method based on user cultural tourism demand analysis provided in the embodiments of the present invention (as described below).

[0065] Optionally, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor may be a general-purpose processor, including a Central Processing Unit (CPU), Network Processor (NP), System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0066] and, Figure 1 The structure shown is for illustrative purposes only. The route planning system based on user cultural tourism demand analysis may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices.

[0067] In one alternative example, the route planning system based on user cultural and tourism demand analysis can be a server with data processing capabilities.

[0068] Combination Figure 2 This invention also provides a route planning method based on user cultural and tourism demand analysis, which can be applied to the aforementioned route planning system based on user cultural and tourism demand analysis. The method steps defined in the process of the route planning method based on user cultural and tourism demand analysis can be implemented by the route planning system based on user cultural and tourism demand analysis (hereinafter referred to as the route planning system). The following will discuss... Figure 2 The specific process shown will be explained in detail.

[0069] Step S110: Obtain the cultural and tourism needs information provided by the target user.

[0070] In this embodiment, the route planning system can obtain user travel needs information provided by the target user. For example, the target user can input information via a mobile phone, computer, or other terminal device to generate the user travel needs information. The terminal device can then send this information to the route planning system, allowing the system to obtain it. In one exemplary implementation scenario, the user travel needs information could be something like, "Planning a trip for two people, hoping for a 3-day, 2-night itinerary, with a preference for historical sites and natural landscapes."

[0071] Step S120: Determine at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information.

[0072] In this embodiment, after obtaining the user's cultural and tourism demand information, the route planning system can determine at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information. Each demand representation information includes one demand sub-information (e.g., "planning a trip for 2 people, hoping for a 3-day, 2-night itinerary, preferring historical sites and natural landscapes" can include three demand sub-information: "planning a trip for 2 people," "hoping for a 3-day, 2-night itinerary," and "preferring historical sites and natural landscapes") and the corresponding emotion-related information. This emotion-related information is obtained by analyzing the target user's provisioning operation for the demand sub-information (i.e., the target user's input operation, through which the demand sub-information is formed, such as inputting "planning a trip for 2 people").

[0073] Step S130: Determine the random semantic vector.

[0074] In this embodiment, the path planning system can also determine a random semantic vector. This random semantic vector is either randomly generated or formed by integrating other cultural and tourism-related information into a randomly generated semantic vector. That is, the random semantic vector can be either a purely random semantic vector or a semantic vector carrying both random semantic information and cultural and tourism-related semantic information, depending on the specific needs. For example, when it is necessary to reduce computational load and improve efficiency, the random semantic vector can be a purely random semantic vector. Alternatively, to avoid introducing excessive noise or interference, the random semantic vector can also be a semantic vector carrying both random semantic information and cultural and tourism-related semantic information.

[0075] Step S140: Semantically encode the random semantic vector at multiple time steps.

[0076] In this embodiment, after obtaining the at least one demand representation information and the random semantic vector, the path planning system can perform semantic encoding on the random semantic vector at multiple time steps. During the semantic encoding process at each time step, the semantic information carried by the at least one demand representation information is fused into the random semantic vector. Thus, through semantic encoding at multiple time steps, the at least one demand representation information can be fully semantically mined, ensuring that the demand-encoded semantic vector formed in the last time step can fully represent the semantic information in the at least one demand representation information. This approach balances the accuracy and richness of semantic representation and better balances the user's tourism needs information provided by the target user with the target user's potential tourism possibilities.

[0077] Step S150: Semantically decode the demand-encoded semantic vector formed by the semantic encoding of the last time step to form the target planning path corresponding to the user's cultural and tourism demand information.

[0078] In this embodiment of the application, after performing semantic encoding at multiple time steps, the path planning system can perform semantic decoding on the demand-encoded semantic vector formed by the semantic encoding of the last time step to form the target planning path corresponding to the user's cultural tourism demand information, such as "Mutianyu Great Wall - Palace Museum in Beijing - Shanji Garden".

[0079] It should be noted that in step S140, semantic encoding can be implemented using a trained encoding neural network model, and in step S150, semantic decoding can be implemented using a trained decoding neural network model. Furthermore, the encoding neural network model and the decoding neural network model can be trained together, for example, by learning from sample tourism demand information and planned route labels. That is, the sample tourism demand information can be processed according to steps S120-S150 to obtain the corresponding planned route. Then, the error between the planned route and the planned route label is calculated, and the model parameters of the encoding neural network model and the decoding neural network model are adjusted and updated in the direction of reducing this error until the error converges. The specific network architecture and semantic decoding process of the decoding neural network model are not the focus of this application. Reference can be made to existing technologies (for example, the semantic vector formed by the semantic encoding of the last time step can be determined as the first vector to be decoded; then, self-attention processing and fully connected processing can be performed on the first vector to be decoded; then, classification functions such as softmax can be used to process the fully connected vector to obtain the corresponding probability distribution, i.e., the probability of each cultural tourism project in the cultural tourism corpus; the cultural tourism project with the highest probability can be determined as the first path planning node in the target planning path; then, the project content corresponding to the first path planning node can be mapped into a vector space, and then the mapped...). The vector is concatenated with the previous (i.e., the first) vector to be decoded to form the second vector to be decoded. Then, the second vector to be decoded is subjected to self-attention processing and fully connected processing. Then, the fully connected vector can be processed by classification functions such as softmax to obtain the corresponding probability distribution, that is, the probability of each cultural and tourism project in the cultural and tourism corpus. The cultural and tourism project with the highest probability is determined as the second path planning node in the target planning path. In this way, each path planning node can be obtained. Finally, the target planning path is formed by combining the path planning nodes according to the generation order of each path planning node. The focus of this application is semantic encoding, that is, how to encode to form a reliable demand encoding semantic vector, the specific process of which is described below.

[0080] Based on the above method, firstly, semantic encoding enables the extraction of latent semantic information from users' cultural and tourism needs, resulting in a target planning path that better matches the actual cultural and tourism needs of the target users. Secondly, the use of random semantic vectors during semantic encoding avoids overfitting and allows for a better balance between the user's cultural and tourism needs and potential possibilities during semantic decoding, thus generating a more accurate and high-quality path. Thirdly, semantic encoding not only extracts latent semantic information from the user's cultural and tourism needs themselves but also from the latent semantic information related to the emotions reflected in the target user's actions. This allows the resulting demand-encoded semantic vector to represent the most authentic needs of the target users, further improving the reliability of the generated target planning path. Therefore, this method can address the problem of relatively low reliability in existing technologies for planning cultural and tourism needs.

[0081] In the first part, regarding step S120, it should be noted that the specific method for determining at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information is not limited and can be selected according to actual needs.

[0082] For example, in an alternative implementation, the user's cultural and tourism demand information can be directly used as a demand sub-information, and the duration for which the target user inputs the user's cultural and tourism demand information can be used as an emotion-related information. In this way, the demand sub-information and the emotion-related information can be combined to form a demand representation information.

[0083] For example, in another alternative implementation, in order to ensure that the determined demand representation information can better represent the actual needs of the target user, the above step S120 may further include the following:

[0084] The first step is to segment the user's cultural and tourism demand information based on the similarity between text content, forming at least one demand sub-information. For example, the vectors corresponding to the text can be determined based on the Bag of Words (BoW) model or TF-IDF, and then the similarity between the vectors can be calculated. In this way, the similarity between text content can be obtained. Texts with similarity less than a threshold can be assigned to different demand sub-information. Alternatively, the similarity between text content belonging to the same sentence can be determined as high, and the similarity between text content belonging to different sentences can be determined as low, that is, one sentence is one demand sub-information. Or, the segmentation can also be based on the number of people, time, location, etc.

[0085] The second step involves determining the input rate (e.g., the ratio of the number of characters or words to the input time) for each sub-information of the requirement, based on the number of characters or words and the input duration of the sub-information. It also involves determining the input stability (e.g., the ratio of the number of characters to the number of deleted characters, or the ratio of the number of words to the number of deleted words) based on the number of characters or words and the number of characters or words deleted during the input process. Finally, it involves determining the distribution dispersion (e.g., the dispersion of each input interval) based on the input interval duration between any two adjacent characters or words in the sub-information. It should be noted that a higher input rate can, to some extent, indicate relatively stable emotions and a relatively positive attitude; higher input stability can also, to some extent, indicate relatively stable emotions; and higher distribution dispersion can, to some extent, indicate relatively unstable emotions, greater emotional fluctuations, or indecisiveness.

[0086] The third step is to determine the input rate, input stability, and distribution dispersion of the demand sub-information as the emotion-related information corresponding to the demand sub-information, and to construct corresponding demand representation information based on the demand sub-information and the emotion-related information corresponding to the demand sub-information. In this way, at least one demand representation information can be formed.

[0087] In the second part, regarding step S130, it should be noted that the specific method for determining the random semantic vector is not restricted and can be selected according to actual needs.

[0088] For example, in an alternative implementation, a random vector can be directly generated and determined as the random semantic vector.

[0089] For example, in another alternative implementation, to avoid introducing excessive interference or noise that could lead to low semantic accuracy in subsequent semantic encoding, step S130 may further include the following (in conjunction with...). Figure 3 ):

[0090] The first step is to randomly generate an initial semantic vector; for example, the initial semantic vector can be Gaussian noise. In addition, in order to avoid the problem of noise shift, the mean of each parameter in the initial semantic vector can be equal to 0.

[0091] The second step is to perform vector space mapping on the cultural and tourism corpus to form a cultural and tourism corpus mapping vector. The cultural and tourism corpus includes multiple cultural and tourism projects (such as scenic spots) and the project content of each cultural and tourism project (such as the introduction and description of the scenic spot). Each path planning node in the target planning path belongs to one of the multiple cultural and tourism projects.

[0092] The third step involves performing matrix multiplication between the initial semantic vector and the cultural tourism corpus mapping vector to form a relevance parameter distribution, and then applying this distribution based on the target masking matrix (in...). Figure 3 In the diagram (where black represents "0" and white represents "1"), the distribution of the relevant parameters is partially masked to form a masked parameter distribution. The target masking matrix is ​​formed during the learning process of the corresponding neural network model through sample cultural tourism demand information and planned route labels. In this way, the unimportant parameters in the distribution of the relevant parameters can be hidden through the target masking matrix, so that the main parameters can be focused on.

[0093] The fourth step involves filtering the occlusion parameter distribution by the maximum value of columns and rows or by the mean of columns and rows, forming a first filtering distribution and a second filtering distribution. For example, the maximum value or mean of each column in the occlusion parameter distribution can be determined to form the first filtering distribution, and the maximum value or mean of each row in the occlusion parameter distribution can be determined to form the second filtering distribution, i.e., the most similar relationship in the row direction and column direction can be mined respectively.

[0094] The fifth step involves fusing the first and second screening distributions to form a random semantic vector. For example, the first and second screening distributions (transposes) can be concatenated or added to form the random semantic vector. Based on this, random noise can be represented by the random semantic vector, and cultural and tourism corpus is also incorporated. This allows for the representation or fitting of potential demands through random noise in subsequent semantic encoding, and the encoding of cultural and tourism semantics can be guided by cultural and tourism corpus in subsequent semantic encoding.

[0095] The third part, regarding step S140, should be noted that the specific method of semantically encoding the random semantic vector at multiple time steps is not limited and can be selected according to actual needs.

[0096] For example, in an alternative implementation, at the first time step, the random semantic vector and the mapping vector corresponding to at least one demand representation information can be fused (e.g., through cross-attention processing) to obtain the demand-encoded semantic vector for the first time step. In the second time step and each subsequent time step, the demand-encoded semantic vector of the previous time step and the mapping vector corresponding to at least one demand representation information can be fused (e.g., through cross-attention processing) to obtain the demand-encoded semantic vector for the corresponding time step.

[0097] For example, in another alternative implementation, in order to ensure that the semantic information in the at least one requirement representation information is fully extracted through semantic encoding at the multiple time steps, step S140 may further include steps S141, S142, S143, S144, and S145, the specific contents of which are as follows (in conjunction with...). Figure 4 ).

[0098] Step S141: Based on at least one demand sub-information, semantically encode the first local vector in the random semantic vector to form at least one demand encoding vector of the first local vector.

[0099] In this embodiment, semantic encoding can be performed on a first local vector in the random semantic vector based on at least one demand sub-information (i.e., each demand sub-information included in the at least one demand representation information) to form at least one demand encoding vector of the first local vector. For example, the first demand sub-information can be fused into the first local vector to form a first demand encoding vector corresponding to the first local vector, and the second demand sub-information can be fused into the first local vector to form a second demand encoding vector corresponding to the first local vector.

[0100] Step S142: Based on at least one demand encoding vector and at least one emotion-related information of the first local vector, determine the emotion encoding vector of the first local vector.

[0101] In this embodiment, after forming at least one demand encoding vector of the first local vector, the emotion encoding vector of the first local vector can be determined based on at least one demand encoding vector and at least one emotion-related information. That is, at least one emotion-related information can be fused into at least one demand encoding vector to form an emotion encoding vector carrying emotion-related semantics. Furthermore, it should be noted that since the semantic information of the random semantic vector and the demand sub-information is closer in semantic space, they can be fused first. Then, based on this fusion, since the semantic information of the demand sub-information and the emotion-related information are also closer in semantic space, further fusion can be performed. This progressive fusion avoids the problem of poor fusion results.

[0102] Step S143: Based on the at least one demand sub-information, semantically encode the second local vector in the random semantic vector to form at least one demand encoding vector of the second local vector.

[0103] In this embodiment, based on the at least one demand sub-information, the second local vector in the random semantic vector can be semantically encoded (the specific implementation process can be the same as in step S141) to form at least one demand encoding vector of the second local vector. For example, the first demand sub-information can be fused into the second local vector to form the first demand encoding vector corresponding to the second local vector, and the second demand sub-information can be fused into the second local vector to form the second demand encoding vector corresponding to the second local vector. The first local vector and the second local vector partially overlap, and both belong to the random semantic vector. Therefore, since the first local vector and the second local vector are semantically encoded separately, semantic fusion of the random semantic vector can be achieved from the perspective of different vector distribution positions, resulting in higher fusion accuracy. Furthermore, since the first local vector and the second local vector partially overlap, the correlation between the sentiment encoding vector of the first local vector and the sentiment encoding vector of the second local vector is higher. Therefore, during fusion, it can be ensured that the various parts of the formed demand encoding semantic vector are not completely isolated, facilitating effective representation of global semantic information.

[0104] Step S144: Based on at least one demand encoding vector of the second local vector and the at least one emotion-related information, determine the emotion encoding vector of the second local vector.

[0105] In this embodiment of the application, after forming at least one demand encoding vector of the second local vector, the emotion encoding vector of the second local vector can be determined based on at least one demand encoding vector of the second local vector and the at least one emotion-related information. The specific implementation method can be the same as the implementation method of step S142 described above.

[0106] Step S145: Based on the part of the emotion encoding vector of the first local vector and the part of the emotion encoding vector of the second local vector that does not overlap with the first local vector, the demand encoding semantic vector of the last time step is formed.

[0107] In this embodiment, after forming the emotion encoding vectors of the first local vector and the second local vector, the demand encoding semantic vector for the last time step can be formed based on the vectors in the emotion encoding vectors of the first and second local vectors that do not overlap with the first local vector. For example, the vectors in the emotion encoding vectors of the first and second local vectors that do not overlap with the first local vector can be concatenated to form the demand encoding semantic vector for the last time step. That is, in the first time step, the emotion encoding vector of the first local vector can be obtained; in the second time step, the emotion encoding vector of the second local vector can be obtained; and in the last time step, the demand encoding semantic vector for the last time step can be obtained. Alternatively, in other implementations, a third local vector may be included that partially overlaps with the second local vector. In this way, the emotion encoding vector of the third local vector can be obtained in the third time step. Then, in the last time step, the emotion encoding vector of the first local vector, the part of the emotion encoding vector of the second local vector that does not overlap with the first local vector, and the part of the emotion encoding vector of the third local vector that does not overlap with the second local vector can be concatenated to form the demand encoding semantic vector of the last time step.

[0108] Alternatively, in step S141 above, the specific method of semantically encoding the first local vector in the random semantic vector is not limited. For example, in an alternative implementation, in order to achieve sufficient fusion of demand sub-information, i.e., to ensure the reliability of the semantic encoding of demand sub-information, while avoiding overfitting due to excessive fusion depth, step S141 above can further include the following steps S141a, S141b, S141c, and S141d, combined with... Figure 5The specific details of each step are as follows (the following content only applies to any one requirement sub-information; the semantic encoding method for other requirement sub-information can be the same).

[0109] Step S141a: Perform vector space mapping on any one of the demand sub-information to form a demand mapping vector corresponding to the one demand sub-information.

[0110] In this embodiment, any sub-information of demand can be mapped into a vector space to form a demand mapping vector corresponding to that sub-information. Specifically, vector space mapping can be implemented using a word embedding model. For example, for a sub-information of demand "preferring historical sites and natural landscapes", word segmentation and embedding can yield:

[0111] Tendency: [0.1, -0.3, 0.3, 0.7, 0.5, 0.6, 0.4, 0.8, 0.9, ..., 1.1];

[0112] Historical values: [0.2, 0.1, 0.1, 0, 0.4, 0.7, 0.6, 0.1, 0.8, ..., 0.1];

[0113] Remains: [-0.1,0.2,0.1,-0.2,0.3,0.5,0.5,0.6,-0.7,...,0.2];

[0114] And: [0.3,0.4,-0.5,0.2,0.7,0.5,0.9,1.0,0.1,...,1.2];

[0115] Nature: [0.4,0.1,0.6,0.3,0.8,0.8,1.0,0.1,0.2,...,0.3];

[0116] Landscape: [0.5,0.6,0.1,0.2,0.6,1.0,0.1,0.2,1.3,...,0.4].

[0117] Step S141b: In the first fusion stage, based on the first local vector in the random semantic vector, cross-attention processing is performed on the demand mapping vector corresponding to any demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the first fusion stage.

[0118] In this embodiment, in the first fusion stage, cross-attention processing can be performed on the demand mapping vector corresponding to any one of the demand sub-information based on the first local vector in the random semantic vector to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector (e.g., by addition), the connection result is locally deactivated (for example, the first local deactivation matrix in the corresponding neural network model can be multiplied bitwise with the connection result, and the first local deactivation matrix can be composed of two parameters, "0" and "1") to form the fusion output vector of the first fusion stage.

[0119] In step S141c, in the second fusion stage, based on the fusion output vector of the first fusion stage, cross-attention processing is performed on the demand mapping vector corresponding to any one of the demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the second fusion stage.

[0120] In the embodiments of this application, in the second fusion stage, cross-attention processing can be performed on the demand mapping vector corresponding to any one of the demand sub-information based on the fusion output vector of the first (i.e. the previous) fusion stage to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector (such as by addition), the connection result is locally deactivated (for example, the second local deactivation matrix in the corresponding neural network model can be multiplied bitwise with the connection result) to form the fusion output vector of the second fusion stage.

[0121] Step S141d: The fusion output vector of the last fusion stage is determined as a requirement encoding vector of the first local vector.

[0122] In this embodiment, through the aforementioned progressive fusion, after obtaining the fusion output vector of the last fusion stage, the fusion output vector of the last fusion stage can be determined as a requirement encoding vector of the first local vector. It should be noted that if there are only two fusion stages, the fusion output vector of the second fusion stage can be determined as a requirement encoding vector of the first local vector. If there are three fusion stages, the fusion output vector of the third fusion stage can be determined as a requirement encoding vector of the first local vector.

[0123] Optionally, in step S142 above, the specific method for determining the emotion encoding vector of the first local vector is not limited. For example, in an alternative implementation, to balance the reliability and efficiency of fusing emotion-related information, step S142 may further include steps S142a, S142b, S142c, S142d, and S142e. The specific content of each step is as follows, combined with... Figure 6 As shown.

[0124] Step S142a: Perform vector space mapping on each emotion-related information to form an emotion mapping vector corresponding to each emotion-related information.

[0125] In this embodiment of the application, each emotion-related information can be mapped into a vector space to form an emotion mapping vector corresponding to each emotion-related information. For example, the vector space mapping of emotion-related information can be achieved through a word embedding model.

[0126] Step S142b: In at least one demand encoding vector of the first local vector, based on the correspondence between the demand sub-information and emotion-related information included in the same demand representation information, the demand encoding vector corresponding to each of the emotion mapping vectors is determined respectively.

[0127] In this embodiment, after obtaining the emotion mapping vector, the demand encoding vector corresponding to each emotion mapping vector can be determined based on the correspondence between the demand sub-information and emotion-related information included in the same demand representation information, within at least one demand encoding vector of the first local vector. For example, if there is a first demand representation information and a second demand representation information, a correspondence can be established between the demand encoding vector corresponding to the demand sub-information in the first demand representation information and the emotion mapping vector corresponding to the emotion-related information in the first demand representation information; similarly, a correspondence can be established between the demand encoding vector corresponding to the demand sub-information in the second demand representation information and the emotion mapping vector corresponding to the emotion-related information in the second demand representation information, to ensure the reliability of subsequent fusion.

[0128] Step S142c: For each of the emotion mapping vectors, the emotion mapping vector and the corresponding demand encoding vector are interactively gating to form a first gating fusion vector and a second gating fusion vector.

[0129] In this embodiment of the application, after determining the correspondence, for each of the emotion mapping vectors, the emotion mapping vector and the corresponding demand encoding vector can be interactively gating to form a first gating fusion vector and a second gating fusion vector.

[0130] Step S142d: Connect the first gating fusion vector, the second gating fusion vector, the emotion mapping vector, and the demand encoding vector to form a corresponding connection result.

[0131] In this embodiment of the application, after obtaining the first gating fusion vector and the second gating fusion vector, considering that gating adjustment can be more efficient than attention processing, but requires high quality and stability of the corresponding vectors during fusion, poor quality may lead to poor fusion effect. Therefore, the first gating fusion vector, the second gating fusion vector, the emotion mapping vector and the demand encoding vector can be connected to form the corresponding connection result. In this way, the semantic loss problem caused by gating fusion can be avoided.

[0132] Step S142e: When there are multiple emotion mapping vectors, the connection results corresponding to the multiple emotion mapping vectors are added together to form the emotion encoding vector of the first local vector.

[0133] In this embodiment, after obtaining the connection result corresponding to each emotion mapping vector, when there are multiple emotion mapping vectors, the connection results corresponding to the multiple emotion mapping vectors are added together to form the emotion encoding vector of the first local vector. When there is only one emotion mapping vector, the connection result corresponding to that emotion mapping vector can be used as the emotion encoding vector.

[0134] Alternatively, in step S142c above, the specific method of gating the interaction between the emotion mapping vector and the corresponding demand encoding vector is not limited. For example, in an alternative implementation, in order to fully fit the potential linear and nonlinear relationships in the gating process to improve the semantic representation ability between the formed first gating fusion vector and the second gating fusion vector, step S142c above may further include steps c1, c2, c3 and c4, the specific contents of each step are as follows.

[0135] Step c1: Perform linear mapping and nonlinear processing on the emotion mapping vector to form the first gating parameter.

[0136] In this embodiment of the application, the emotion mapping vector can be linearly mapped and nonlinearly processed (i.e., first linearly mapped, and then nonlinearly processed on the result of the linear mapping) to form the first gating parameter.

[0137] Step c2 involves performing linear mapping and nonlinear processing on the demand encoding vector corresponding to the emotion mapping vector to form the second gating parameter.

[0138] In this embodiment of the application, the demand encoding vector corresponding to the emotion mapping vector can be linearly mapped and nonlinearly processed (i.e., linear mapping is performed first, and then nonlinear processing is performed on the result of the linear mapping) to form a second gating parameter.

[0139] Step c3: Multiply the parameters at corresponding positions between the first gating parameter and the demand encoding vector corresponding to the emotion mapping vector to form the first gating fusion vector.

[0140] In this embodiment, after forming the first gating parameter, the parameters at corresponding positions between the first gating parameter and the demand encoding vector corresponding to the emotion mapping vector can be multiplied (i.e., multiplied bitwise) to form a first gating fusion vector. In this way, the parameters at important positions in the first gating parameter corresponding to their positions in the demand encoding vector can receive focused attention, thus achieving gating adjustment of the demand encoding vector.

[0141] Step c4: Multiply the parameters at corresponding positions between the second gating parameter and the emotion mapping vector to form the second gating fusion vector.

[0142] In this embodiment of the application, after the second gating parameter is formed, the parameters at corresponding positions between the second gating parameter and the emotion mapping vector can be multiplied (i.e., multiplied bitwise) to form the second gating fusion vector.

[0143] Alternatively, in step c1 above, the specific method of linear mapping and nonlinear processing of the emotion mapping vector is not limited. For example, in an alternative implementation, in order to ensure that the formed first gating parameter can fully represent (i.e. fully extract) important semantic information, step c1 above may further include the following:

[0144] The first step is to multiply the emotion mapping vector and its transpose to form an emotion attention distribution. Then, based on the emotion attention distribution, the emotion mapping vector is weighted and summed to form an emotion attention vector. Based on this, self-attention processing of the emotion mapping vector can be achieved, that is, the important semantics inside can be extracted.

[0145] The second step involves linearly mapping the emotion attention vector based on the target weight matrix and the target bias parameter (both parameters are trained in the corresponding neural network model) to form an emotion linear vector (e.g., emotion linear vector = target weight matrix * emotion attention vector + target bias parameter). Additionally, a nonlinear mapping can be performed on the emotion linear vector based on a target nonlinear function (such as the sigmoid function) to form an emotion nonlinear vector.

[0146] The third step is to determine the emotion nonlinear vector as the first gating parameter.

[0147] Further explanation is needed for step S140 above. In order to effectively determine multiple local vectors (such as the first local vector and the second local vector), step S140 may further include the following:

[0148] The first step is to determine the target segmentation box (e.g., m*n), the target movement direction (e.g., from left to right), the target segmentation scale (e.g., 10, 20, 40, etc.), and the original vector coordinates for local vector segmentation in the random semantic vector (e.g., the first position of the target segmentation box coincides with the first position in the random semantic vector).

[0149] The second step is to determine the first local vector when the target segmentation box is located at the original vector coordinates.

[0150] Third, in the random semantic vector, starting from the original vector coordinates, the target segmentation box is moved a distance represented by the target segmentation scale along the target movement direction, and then the local vector currently enclosed by the target segmentation box in the random semantic vector is determined as the second local vector.

[0151] In summary, the path planning method and system based on user cultural tourism demand analysis provided by this invention first obtains user cultural tourism demand information provided by the target user; second, it determines at least one demand representation information corresponding to at least one demand sub-information included in the user cultural tourism demand information; then, it determines a random semantic vector; further, it performs semantic encoding on the random semantic vector at multiple time steps, wherein, in the semantic encoding process of each time step, the semantic information carried by at least one demand representation information is integrated into the random semantic vector; finally, it performs semantic decoding on the demand-encoded semantic vector formed by the semantic encoding of the last time step to form the target planning path corresponding to the user cultural tourism demand information. Based on the above method, firstly, because semantic encoding is performed, the potential semantic information in the user cultural tourism demand information can be mined, thereby obtaining a target planning path that is more closely matched to the actual cultural tourism demand of the target user. Secondly, because random semantic vectors are used in the semantic encoding process, overfitting can be avoided in the semantic encoding process, and the user cultural tourism demand information and potential possibilities provided by the target user can be better balanced in the semantic decoding process, thereby generating a more accurate and high-quality path. Thirdly, because the semantic encoding process not only mines the latent semantic information of the user's cultural and tourism needs themselves, but also mines the latent semantic information related to the emotions reflected in the target user's actions, the resulting demand-encoded semantic vector can represent the target user's most authentic needs, thereby further improving the reliability of the generated target planning path. Therefore, it can improve the problem of relatively low reliability in the path planning of cultural and tourism needs in existing technologies.

[0152] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0153] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A route planning method based on user cultural tourism demand analysis, characterized in that, include: Obtain information on the cultural and tourism needs of target users; Determine at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information, wherein each demand representation information includes a demand sub-information and an emotion-related information corresponding to the demand sub-information, and the emotion-related information is obtained by analyzing the provision operations performed by the target user for the demand sub-information; An initial semantic vector is randomly generated; the cultural and tourism corpus is mapped into a vector space to form a cultural and tourism corpus mapping vector, wherein the cultural and tourism corpus includes multiple cultural and tourism projects and the content of each cultural and tourism project; the initial semantic vector and the cultural and tourism corpus mapping vector are multiplied by a matrix to form a relevance parameter distribution, and the relevance parameter distribution is locally masked based on a target masking matrix to form a masking parameter distribution, wherein the target masking matrix is ​​formed during the learning process of the corresponding neural network model through sample cultural and tourism demand information and planned path labels; the masking parameter distribution is filtered by the maximum value of columns and rows or the mean of columns and rows to form a first filtering distribution and a second filtering distribution; the first filtering distribution and the second filtering distribution are fused to form a random semantic vector; The random semantic vector is semantically encoded over multiple time steps, including: semantically encoding a first local vector in the random semantic vector based on at least one demand sub-information to form at least one demand encoding vector of the first local vector; determining an emotion encoding vector of the first local vector based on at least one demand encoding vector of the first local vector and at least one emotion-related information; semantically encoding a second local vector in the random semantic vector based on the at least one demand sub-information to form at least one demand encoding vector of the second local vector, wherein the first local vector and the second local vector partially overlap, and both the first local vector and the second local vector belong to a portion of the random semantic vector; determining an emotion encoding vector of the second local vector based on at least one demand encoding vector of the second local vector and the at least one emotion-related information; and forming a demand encoding semantic vector for the last time step based on the portion of the emotion encoding vector of the first local vector and the portion of the emotion encoding vector of the second local vector that does not overlap with the first local vector. The semantic vector of demand encoding formed by the semantic encoding of the last time step is semantically decoded to form the target planning path corresponding to the user's cultural and tourism demand information.

2. The route planning method based on user cultural tourism demand analysis as described in claim 1, characterized in that, The step of determining at least one demand representation information corresponding to at least one demand sub-information included in the user's cultural and tourism demand information includes: Based on the similarity between text content, the user's cultural and tourism demand information is segmented to form at least one demand sub-information. For each of the aforementioned demand sub-information, the input rate of the demand sub-information is determined based on the number of characters or words in the demand sub-information and the input duration of the demand sub-information; the input stability of the demand sub-information is determined based on the number of characters or words in the demand sub-information and the number of characters or words deleted during the input process; and the distribution dispersion is determined based on the input interval duration between each pair of adjacent characters or words in the demand sub-information. The input rate, input stability, and distribution dispersion of the demand sub-information are determined as the emotion-related information corresponding to the demand sub-information. Based on the demand sub-information and the emotion-related information corresponding to the demand sub-information, a corresponding demand representation information is constructed.

3. The route planning method based on user cultural tourism demand analysis as described in claim 1, characterized in that, The step of semantically encoding a first local vector in the random semantic vector based on at least one demand sub-information to form at least one demand-encoded vector of the first local vector includes: Perform vector space mapping on any one of the demand sub-informations to form a demand mapping vector corresponding to that one demand sub-information; In the first fusion stage, based on the first local vector in the random semantic vector, cross-attention processing is performed on the demand mapping vector corresponding to any demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the first fusion stage. In the second fusion stage, based on the fusion output vector of the first fusion stage, cross-attention processing is performed on the demand mapping vector corresponding to any one of the demand sub-information to form a random demand fusion vector. After connecting the random demand fusion vector and the demand mapping vector, the connection result is locally deactivated to form the fusion output vector of the second fusion stage. The fusion output vector of the last fusion stage is determined as a requirement encoding vector of the first local vector.

4. The route planning method based on user cultural tourism demand analysis as described in claim 1, characterized in that, The step of determining the emotion encoding vector of the first local vector based on at least one demand encoding vector and at least one emotion-related information includes: Each emotion-related information is mapped into a vector space to form an emotion mapping vector corresponding to each emotion-related information. In at least one demand encoding vector of the first local vector, based on the correspondence between the demand sub-information and emotion-related information included in the same demand representation information, the demand encoding vector corresponding to each of the emotion mapping vectors is determined respectively. For each of the aforementioned emotion mapping vectors, the emotion mapping vector and the corresponding demand encoding vector are interactively gated to form a first gated fusion vector and a second gated fusion vector. The first gated fusion vector, the second gated fusion vector, the emotion mapping vector, and the demand encoding vector are connected to form the corresponding connection result; When there are multiple emotion mapping vectors, the connection results corresponding to the multiple emotion mapping vectors are added together to form the emotion encoding vector of the first local vector.

5. The route planning method based on user cultural tourism demand analysis as described in claim 4, characterized in that, The step of interactively gating the emotion mapping vector and its corresponding demand encoding vector to form a first gated fusion vector and a second gated fusion vector for each emotion mapping vector includes: The emotion mapping vector is subjected to linear mapping and nonlinear processing to form the first gating parameter; The demand encoding vector corresponding to the emotion mapping vector is linearly mapped and nonlinearly processed to form the second gating parameter; The parameters at corresponding positions between the first gating parameter and the demand encoding vector corresponding to the emotion mapping vector are multiplied to form the first gating fusion vector; The second gating parameter and the corresponding parameters between the emotion mapping vector are multiplied to form the second gating fusion vector.

6. The route planning method based on user cultural tourism demand analysis as described in claim 5, characterized in that, The step of performing linear mapping and nonlinear processing on the emotion mapping vector to form the first gating parameter includes: The emotion mapping vector and its transpose are multiplied to form an emotion attention distribution. Based on the emotion attention distribution, the emotion mapping vector is weighted and summed to form an emotion attention vector. Based on the target weight matrix and the target bias parameter, the emotion attention vector is linearly mapped to form an emotion linear vector; and based on the target nonlinear function, the emotion linear vector is nonlinearly mapped to form an emotion nonlinear vector. The emotion nonlinear vector is determined as the first gating parameter.

7. The route planning method based on user cultural tourism demand analysis as described in claim 1, characterized in that, The step of semantically encoding the random semantic vector at multiple time steps further includes: The target segmentation bounding box, target movement direction, target segmentation scale, and original vector coordinates for local vector segmentation in the random semantic vector are determined. When the target segmentation box is located at the original vector coordinates, the local vector currently enclosed by the target segmentation box in the random semantic vector is determined as the first local vector; In the random semantic vector, starting from the original vector coordinates, the target segmentation box is moved a distance represented by the target segmentation scale along the target movement direction. Then, the local vector currently enclosed by the target segmentation box in the random semantic vector is determined as the second local vector.

8. A route planning system based on user cultural tourism demand analysis, characterized in that, include: A processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the path planning method based on user cultural tourism demand analysis as described in any one of claims 1-7.

Images