A method and system for constructing travel user profiles
By constructing a multimodal sentiment analysis neural network and introducing a decay factor, the tag weights in the user profile are dynamically adjusted, solving the problem that users have difficulty accurately finding attractions of interest in existing technologies. This achieves accurate and dynamic updates to the user profile, improving the accuracy of recommendations and user satisfaction.
Patent Information
- Application Number
- CN202410122882.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-29
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-01-29
AI Technical Summary
Existing methods for building travel user profiles fail to fully utilize multimodal data, making it difficult for users to accurately find attractions and destinations of interest.
By collecting user information, an initial user profile is constructed. Then, a multimodal sentiment analysis neural network is used in conjunction with a decay factor to dynamically adjust the label weights in the user profile, including text data, image data, and interaction data. A forgetting curve is introduced to respond to changes in user interests.
It achieves accurate and dynamic updates of user profiles, improves the accuracy of recommendations and user satisfaction, and avoids the problem of excessive weighting caused by a large number of similar behaviors in a short period of time.
Smart Images

Figure CN117972204B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scenic spot recommendation technology, and in particular to a method and system for constructing travel user profiles. Background Technology
[0002] With the widespread adoption of the internet and mobile devices, the tourism industry has also undergone tremendous changes. More and more people are considering the services offered by online travel platforms when choosing travel destinations, hotels, or other travel-related services. A growing number of businesses are recognizing the importance of building accurate user profiles, precisely targeting user needs, and providing personalized solutions and targeted marketing to different users. Online travel platforms such as Ctrip and Qunar provide users with a wealth of travel information, user reviews, and travel photos, giving users more options when choosing tourist attractions.
[0003] However, this also leads to information overload, making it difficult for users to select travel destinations that truly interest them. Traditional user profiling methods primarily rely on basic user attributes such as age, gender, and occupation, as well as their browsing history and purchasing behavior, often neglecting the dynamic changes in user interests and the varying value of data at different points in time. Users on travel platforms not only leave text reviews but also upload photos and engage in interactive behaviors such as liking and saving. This rich multimodal data is not being utilized, further hindering users' ability to accurately find their desired destinations. Summary of the Invention
[0004] In view of the problems existing in the current travel user profiling system, this invention is proposed.
[0005] Therefore, the problem that this invention aims to solve is that it prevents users from accurately finding the tourist attractions they want.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0007] In a first aspect, embodiments of the present invention provide a method for constructing a travel user profile, which includes the following steps:
[0008] Based on the user information collected by the client APP, the user information is processed through the preprocessing module until the user information data is consistent;
[0009] Based on the user information, construct feature tags for the scenic spots;
[0010] An initial user profile is constructed by combining the user information with the feature tags;
[0011] The initial user profile is input into a multimodal sentiment analysis neural network to obtain the user's sentiment category;
[0012] By analyzing the user information, introducing a decay factor, and combining it with the emotion category, a long-term user profile is obtained.
[0013] As a preferred embodiment of the travel user profile construction method of the present invention, the user information includes text data, image data, interaction data, and static information;
[0014] The text data refers to the comment text posted by the user.
[0015] The image data is the image of a comment posted by a user;
[0016] The interaction data includes search, view, like, favorite, and forward behaviors, as well as the time information of each behavior;
[0017] The static information includes the user's ID, age, gender, occupation, and city of residence;
[0018] The preprocessing module includes the following steps:
[0019] The text data was segmented using the jieba word segmentation tool to remove special characters and stop words;
[0020] The image data is resized and normalized to the range of [0,1], and the image is converted from its original color space to the HSV color space.
[0021] The interactive data is labeled and used as the training and testing sets for the multimodal sentiment analysis neural network;
[0022] The static information is categorized for use in extracting the initial user profile.
[0023] As a preferred embodiment of the travel user profile construction method of the present invention, the method for constructing the feature tags includes:
[0024] By selecting comments from tourist attractions and using the LDA topic model to identify user interests, and combining this with confusion level as an evaluation metric for topic selection, the number of topics is calculated using the following formula:
[0025]
[0026] In the formula, M represents the number of documents, and N represents the number of documents. d p(w) represents the number of words in document d. i |d) represents the probability assigned to the i-th word in document d by the model;
[0027] By combining the similarity between user interests obtained from synonym calculations, synonyms are grouped together.
[0028] To give meaning to a theme, serving as a distinctive label for a tourist attraction.
[0029] As a preferred embodiment of the travel user profile construction method of the present invention, the initial user profile includes the user's basic information, habitual travel scenario information, preferred attraction characteristics, and corresponding initial interest weights.
[0030] The method for constructing the initial user profile includes:
[0031] Extract the user's ID, age, gender, occupation, and city of residence, combine them with the user's travel history, and input them into the database to create an initial user profile;
[0032] The travel history includes seasonal scenarios, weather scenarios, long and short distance scenarios, number of companions scenarios, and cost scenarios;
[0033] The seasonal and weather scenarios can be obtained from the travel time of the user's purchased order. The long / short distance scenario can be obtained by comparing the order destination with the user's long-term residence. Intra-city travel is considered short distance, and inter-city travel is considered long distance. The number of people traveling with the user is determined by whether the user has purchased group tickets. The cost scenario is obtained by the sum of the order costs.
[0034] As a preferred embodiment of the travel user profile construction method of the present invention, the multimodal sentiment analysis neural network includes a text analysis network and an image analysis network;
[0035] The text analysis network consists of a BERT model layer, a dilated convolutional layer, and a Bi-LSTM layer connected in series.
[0036] The image analysis network, ResNet convolutional neural network, is composed of a CBMA attention mechanism.
[0037] The multimodal sentiment analysis neural network uses cross-modal attention to combine text and image data, and uses a Dense layer for sentiment classification. The calculation formula for cross-modal attention is as follows:
[0038]
[0039] In the formula, attn represents the attention score of the i-th image, σ represents the sigmoid function, W represents the trainable matrix, and r v Represented as the feature vector of the image, r t denoted as the feature vector of the text, b represents the bias vector;
[0040] The loss function used in the multimodal sentiment analysis neural network includes:
[0041]
[0042] In the formula, M represents the total number of samples in the training set, and N represents the total number of categories. Let y be the predicted probability value of the i-th sample of the m-th sample. mi This represents the true label of the m-th sample;
[0043] The emotion categories include very disappointed, somewhat disappointed, neutral, fairly good, and very good.
[0044] As a preferred embodiment of the travel user profile construction method of the present invention, the method for obtaining the short-term user profile includes:
[0045] Construct a time window and select the feature label weight change value;
[0046] Determine whether a user has interacted with a certain attraction's feature tags over a long period of time. If the determination is no, no action is taken. If the determination is yes, the weight of the relevant feature tags is moved to the long-term interest tags. The calculation formula is as follows:
[0047]
[0048] In the formula, i represents the user's ID, j represents the user profile's tag number, and W... i,j W represents the updated weight after this interaction. origini,j Let T represent the weight of user i's label before this calculation. i,j This represents the time interval between the last interaction and the current interaction for user i's tag j. Let n represent the basic weight change value of user i's label j in this interaction. i,j This represents the number of times the label j of user i changes within the time window, and S is a constant.
[0049] As a preferred embodiment of the travel user profile construction method of the present invention, the method for obtaining the long-term user profile includes:
[0050] Determine if there are text reviews for attractions; if there are only ratings and no text, update the tag weights based on the ratings.
[0051] When there is text, the tag weight is modified based on the sentiment of the comment and the points of interest mentioned in the comment;
[0052] When a user submits multiple text messages, the one with the highest sentiment bias is used as the sentiment reference. The calculation formula is as follows:
[0053] W i,j =w origimi,j *e(-T i,j / S)+ΔW i,j *rat+ΔW i,j *emo*[j∈rev - them]
[0054] In the formula, ΔW i,j Let represent the basic weight change value of user i's tag j in this comment, rat represent the user's rating of the attraction, emo represent the score corresponding to the sentiment category of the comment, represent the set of attraction features involved in the user's comment, and S represent a constant.
[0055] The methods for determining the value of [j∈rev_them] include:
[0056] When label j belongs to rev_them, [j∈rev_them] takes the value 1;
[0057] When label j does not belong to rev_them, [j∈rev_them] takes the value 0.
[0058] Secondly, embodiments of the present invention provide a travel user profile building system, which includes: a data storage module, a data analysis module, a calculation and processing module, and an output module;
[0059] The data storage module is used to collect and save customer information, and can also store feature tags for attractions.
[0060] The data analysis module is used to classify customer information, and in conjunction with the data storage module, it enables partitioned storage of customer information of different categories.
[0061] The calculation and processing module is used to calculate the feature tags of customer information and attractions, and to display attractions that customers are interested in to customers.
[0062] The output module is used to organize the data and output it uniformly to the client APP used by the customer.
[0063] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described method for constructing a travel user profile.
[0064] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described method for constructing a travel user profile.
[0065] The beneficial effects of this invention are as follows: It comprehensively considers user interaction data, text comments, image comments, and ratings, and uses deep learning and attention mechanisms to extract sentiment tendencies from comprehensive text and image information, dynamically adjusting the tag weights in the user profile; it divides user interest weights into long-term interests and short-term interests, and introduces a forgetting curve as a time decay factor to ensure that the user profile can respond to changes in user interests in a timely manner, improving the accuracy of recommendations and user satisfaction; this method not only makes full use of rich user data, but also effectively avoids the problem of excessive weight increase caused by a large number of similar behaviors in a short period of time through non-linear weight adjustment and time decay strategies, achieving accurate and dynamic updates of the user profile. Attached Figure Description
[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:
[0067] Figure 1 A flowchart illustrating the method for building user profiles for travelers.
[0068] Figure 2 A schematic diagram of a user profiling system for constructing user profiles for travelers. Detailed Implementation
[0069] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0072] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.
[0073] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0074] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0075] Example 1
[0076] Reference Figures 1-2 This is the first embodiment of the present invention, which provides a method for constructing a travel user profile, including the following steps:
[0077] S1. Collect user information through the client APP and organize the user information through the preprocessing module until the user information data is consistent.
[0078] User information includes text data, image data, interaction data, and static information;
[0079] The text data consists of user-posted comments.
[0080] The image data consists of images from user-posted comments;
[0081] Interaction data includes search, view, like, favorite, and share behaviors, as well as the time information for each behavior;
[0082] Static information includes the user's ID, age, gender, occupation, and city of residence;
[0083] The preprocessing module includes the following steps:
[0084] Perform word segmentation on the text data using the jieba word segmentation tool, and remove special symbols and stop words;
[0085] Adjust the size of the image data, normalize the image data to the range of [0, 1], and uniformly convert the images from their original color space to the HSV space;
[0086] Label the interaction data for use as the training set and test set of the multi-modal sentiment analysis neural network;
[0087] Classify the data of static information for extracting the initial user portrait.
[0088] S2. Construct feature tags of scenic spots according to user information.
[0089] Specifically, the method for constructing the feature tags of the scenic spot includes:
[0090] Select partial high-popularity comments from the scenic spot, mine the interest points that users pay attention to therefrom as the feature tags of the scenic spot according to the LDA topic model, and use the perplexity Perplexity as the evaluation index for topic mining to select a suitable topic, with the calculation formula as follows:
[0091]
[0092] In the formula, M represents the number of documents, Nd represents the number of words in document d, and p(wi|d) represents the probability assigned by the model to the i-th word in document d;
[0093] Calculate the similarity between the interest points that users pay attention to obtained by combining synonymous words, and group the synonyms into one group;
[0094] Assign meanings to the topics to serve as the feature tags of the scenic spot.
[0095] S3. Construct an initial user portrait in combination with the feature tags of the scenic spot.
[0096] Specifically, the initial user portrait includes the user's basic information, habitual travel scenario information, preferred scenic spot features, and the corresponding initial interest weight:
[0097] The construction method of the initial user portrait includes:
[0098] Extract the user's ID, age, gender, occupation and usual residence city, combine the user's travel history, and input them into the database for use as the initial user portrait;
[0099] The travel history includes season scenarios, weather scenarios, long / short trip scenarios, companion number scenarios, and cost scenarios;
[0100] Among them, seasonal and weather scenarios can be obtained from the travel time of the user's purchased order; long and short distance scenarios can be obtained by comparing the order destination with the user's long-term residence; intra-city travel is considered short distance, and inter-city travel is considered long distance; the number of people traveling in a group is determined by whether the user purchased group tickets; and the cost scenario is obtained by the sum of the order costs.
[0101] S4. Input the initial user profile into the multimodal sentiment analysis neural network to obtain the user's sentiment category.
[0102] Multimodal sentiment analysis neural networks include text analysis networks and image analysis networks;
[0103] The text analysis network consists of a BERT model layer, a dilated convolutional layer, and a Bi-LSTM layer connected in series.
[0104] The image analysis network ResNet is composed of a convolutional neural network combined with the CBMA attention mechanism;
[0105] Multimodal sentiment analysis neural networks use cross-modal attention to combine text and image data, and use Dense layers for sentiment classification. The formula for calculating cross-modal attention is as follows:
[0106]
[0107] In the formula, attn represents the attention score of the i-th image, σ represents the sigmoid function, W represents the trainable matrix, and r v Represented as the feature vector of the image, r t denoted as the feature vector of the text, b represents the bias vector;
[0108] The loss functions used in multimodal sentiment analysis neural networks include:
[0109]
[0110] In the formula, M represents the total number of samples in the training set, and N represents the total number of categories. Let y be the predicted probability value of the i-th sample of the m-th sample. mi This represents the true label of the m-th sample, using one-hot encoding;
[0111] The emotion categories include very disappointed, somewhat disappointed, neutral, fairly good, and very good.
[0112] S5. Analyze user information, introduce a decay factor, and combine it with emotion categories to obtain long-term user profiles.
[0113] Specifically, methods for obtaining short-term user profiles include:
[0114] S5.1 Construct a time window and select the feature label weight change value:
[0115] The time window is set to 30 minutes, and the basic weight change values corresponding to various operations are determined. In this embodiment, the basic weight change values are selected as follows: the weight value of search is 2, the weight value of view is 1, the weight value of like is 2, the weight value of favorite is 2, and the weight value of forward is 3.
[0116] S5.2 Based on the user's operation, obtain the features of the attractions involved and calculate the change in interest. If the attraction contains features a, b, or c, then modify the weights of tags a, b, or c in the short-term interest profile. The calculation formula is as follows:
[0117]
[0118] In the formula, i represents the user's ID, j represents the user profile's tag number, and Wi... ,j W represents the updated weight after this interaction. origini,j Let T represent the weight of user i's label before this calculation. i,j This represents the time interval between the last interaction and the current interaction for user i's tag j. Let n represent the basic weight change value of user i's label j in this interaction. i,j This represents the number of times the label j of user i changes within the time window, and S represents a constant, which is taken as 5 / 168 in this embodiment.
[0119] Furthermore, methods for obtaining long-term user profiles include:
[0120] Based on the method of obtaining short-term user profiles, the changes of each interest tag are judged. When the number of interactions with an interest tag accounts for more than 50% of the total number of interactions and the time span exceeds 7 days, the weight of the interest tag is moved to the long-term interest weight, while the weight of the tags involved in the short-term interest profile is set to zero.
[0121] Analyze travel history and update the weight of long-term interests based on the sentiment category of user reviews and ratings of attractions, combined with the time decay factor;
[0122] Determine if there are text reviews for attractions; if there are only ratings and no text, update the tag weights based on the ratings.
[0123] When there is text, the tag weight is modified based on the sentiment of the comment and the points of interest mentioned in the comment;
[0124] When a user submits multiple text messages, the one with the highest sentiment bias is used as the sentiment reference. The calculation formula is as follows:
[0125] W i,j=w origimi,j *e(-T i,j / S)+ΔW i,j *rat+ΔW i,j *emo*[j∈rev - them]
[0126] In the formula, ΔW i,j Let represent the basic weight change value of user i's tag j in this comment, rat represent the user's rating of the attraction, emo represent the score corresponding to the sentiment category of the comment, rev_them represent the set of attraction features involved in the user's comment, and S represent a constant, which is 300 / 7 in this embodiment;
[0127] The methods for determining the value of [j∈rev_them] include:
[0128] When label j belongs to rev_them, [j∈rev_them] takes the value 1;
[0129] When label j does not belong to rev_them, [j∈rev_them] takes the value 0.
[0130] Furthermore, by using the forgetting curve as a time decay factor, the timestamps in user activities are not a single response, but rather a capture and reflection of long-term and short-term trends in user behavior through multiple time windows (day, week, and month).
[0131] Based on the frequency and importance of user behavior, multiple time windows are defined. For example, 1 to 3 days is a short-term window to capture short-term user behavior and the impact of sudden events, and the highest frequency of updates is used; 3 days to 1 week is a medium-term window to analyze the seasonal changes and holiday effects of user interests; and 1 week to 1 month is a long-term window to observe and predict the long-term trends of user interests.
[0132] Furthermore, a feature change tracking module is required, and the specific steps include:
[0133] Establish a feature change monitoring mechanism: use time series analysis technology to track changes in user features over time;
[0134] Design a change detection algorithm: combine statistical methods to monitor the degree to which features deviate from their normal range;
[0135] Set thresholds and trigger conditions: Determine the acceptable range of feature changes or thresholds to identify when user behavior changes are significant enough to require updating the profile. Set trigger conditions to detect behaviors that exceed the above thresholds in real time and trigger the user profile update process.
[0136] Feedback loop for feature changes: A feedback loop is created by combining user feedback with the results of system monitoring to continuously calibrate thresholds and improve the accuracy of the detection algorithm;
[0137] The feature change tracking module is used to extract users' repeated viewing of attractions with the same features, preventing the problem of excessive weighting caused by a large number of similar behaviors.
[0138] In summary, this method comprehensively considers user interaction data, text comments, image comments, and ratings. Utilizing deep learning and attention mechanisms, it extracts sentiment trends from the integrated text and image information, dynamically adjusting the tag weights in the user profile. User interest weights are categorized into long-term and short-term interests, and a forgetting curve is introduced as a time decay factor to ensure that the user profile can respond promptly to changes in user interests, improving recommendation accuracy and user satisfaction. This approach not only fully leverages rich user data but also effectively avoids the problem of excessive weight increases caused by a large number of similar behaviors within a short period through non-linear weight adjustment and time decay strategies, achieving accurate and dynamic updates to the user profile.
[0139] Example 2
[0140] Based on the first embodiment, this embodiment further provides a travel user profile building system, including a data storage module, a data analysis module, a computing and processing module, and an output module;
[0141] The data storage module is used to collect and save customer information, and can also store feature tags for attractions.
[0142] The data analysis module is used to classify customer information, and combined with the data storage module, it enables partitioned storage of customer information of different categories.
[0143] The calculation and processing module is used to calculate the feature tags of customer information and attractions, and to display attractions that customers are interested in to customers.
[0144] The output module is used to organize the data and output it uniformly to the client APP used by the customer.
[0145] This embodiment also provides a computer device applicable to the travel user profile construction method, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the travel user profile construction method proposed in the above embodiment.
[0146] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0147] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for constructing travel user profiles as proposed in the above embodiments.
[0148] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0149] Example 3
[0150] Based on the previous two embodiments, this embodiment provides a method for constructing travel user profiles. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.
[0151] Compared to blind push notifications, this embodiment significantly increases the interaction between users and push notifications, ensuring that users can more accurately find the information they want to follow.
[0152] Keyword interaction probability Page dwell time Blindly pushing 17 1.4min Example 1 76 3.1min
[0153] As can be seen from the table above, in this embodiment, the number of times users interacted with the pushed keywords increased significantly, and the overall dwell time became longer, proving that the pushed content made users pay more attention.
[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A travel user profiling method, characterized by: Includes the following steps, User information is collected through a client-side APP and processed through a preprocessing module until the user information data is consistent. Based on the user information, feature tags for the scenic spots are constructed; Based on the aforementioned feature tags, an initial user profile is constructed; The initial user profile is input into a multimodal sentiment analysis neural network to obtain the user's sentiment category; By analyzing the user information, introducing a decay factor, and combining it with the emotion category, a long-term user profile is obtained. The user information includes text data, image data, interactive data, and static information; The text data refers to the comment text posted by the user. The image data is the image of a comment posted by a user; The interaction data includes search, view, like, favorite, and forward behaviors, as well as the time information of each behavior; The static information includes the user's ID, age, gender, occupation, and city of residence; The multimodal sentiment analysis neural network includes a text analysis network and an image analysis network; The text analysis network consists of a BERT model layer, a dilated convolutional layer, and a Bi-LSTM layer connected in series. The image analysis network, ResNet convolutional neural network, is composed of a CBAM attention mechanism. The multimodal sentiment analysis neural network uses cross-modal attention to combine text and image data, and uses a Dense layer for sentiment classification. The calculation formula for cross-modal attention is as follows: ; wherein, denotes the attention score for the i-th image, denotes a sigmoid function, w denotes a trainable matrix, r v denotes a feature vector of the image, r t denotes a feature vector of the text, b denotes a bias vector; The loss function used in the multimodal sentiment analysis neural network includes: ; In the formula, M represents the total number of samples in the training set, N represents the total number of categories, represents the i th prediction probability value of the m th sample, represents the true label of the m th sample; The emotion categories include very disappointed, somewhat disappointed, neutral, fairly good, and very good; The methods for obtaining the long-term user profile include: Determine if there are text reviews for attractions; if there are only ratings and no text, update the tag weights based on the ratings. When there is text, the tag weight is modified based on the sentiment of the comment and the points of interest mentioned in the comment; When a user submits multiple text messages, the one with the highest sentiment bias is used as the sentiment reference. The calculation formula is as follows: ; In the formula, represents the basic weight change value of label j of user i in this comment, rat represents the score of the user to the scenic spot, and emo represents the score corresponding to the comment emotion category, represents the scenic feature set involved in the user comment, and S represents a constant; in, The methods for determining the value include: When label j belongs to hour, The value is 1; When label j does not belong hour, The value is 0.
2. The method for constructing a travel user profile as described in claim 1, characterized in that: The preprocessing module includes the following steps: The text data was segmented using the jieba word segmentation tool to remove special characters and stop words; The image data is resized and normalized to the range of [0,1], and the image is converted from its original color space to the HSV color space. The interactive data is labeled and used as the training and testing sets for the multimodal sentiment analysis neural network; The static information is categorized for use in extracting the initial user profile.
3. The method for constructing a travel user profile as described in claim 2, characterized in that: The method for constructing the feature labels includes: By selecting comments from tourist attractions and using the LDA topic model to identify user interests, and combining this with confusion level as an evaluation metric for topic selection, the number of topics is calculated using the following formula: ; where M represents the number of documents, N d represents the number of words in document d, p(w i represents the probability assigned by the model to the ith word in document d; By combining the similarity between user interests obtained from synonym calculations, synonyms are grouped together. To give meaning to a theme, serving as a distinctive label for a tourist attraction.
4. The method for constructing a travel user profile as described in claim 3, characterized in that: The initial user profile includes the user's basic information, travel habits, preferred attractions, and corresponding initial interest weights: The method for constructing the initial user profile includes: Extract the user's ID, age, gender, occupation, and city of residence, combine them with the user's travel history, and input them into the database to create an initial user profile; The travel history includes seasonal scenarios, weather scenarios, long and short distance scenarios, number of companions scenarios, and cost scenarios; The seasonal and weather scenarios are obtained from the travel time of the user's purchased orders. The long / short distance scenario is obtained by comparing the order destination with the user's long-term residence. Intra-city travel is considered short distance, and inter-city travel is considered long distance. The number of people traveling with the user is determined by whether the user has purchased group tickets. The cost scenario is obtained by the sum of the order costs.
5. The method for constructing a travel user profile as described in claim 4, characterized in that: Methods for obtaining short-term user profiles include: Construct a time window and select the feature label weight change value; Determine whether a user has interacted with a certain attraction's feature tags over a long period of time. If the determination is no, no action is taken. If the determination is yes, the weight of the relevant feature tags is moved to the long-term interest tags. The calculation formula is as follows: ; In the formula, i represents the user's ID, and j represents the tag number of the user profile. This represents the weight updated after this interaction. This represents the weight of user i's label before this calculation. This represents the time interval between the last interaction and the current interaction for user i's tag j. This represents the basic weight change value of user i's label j in this interaction. This represents the number of times the label j of user i changes within the time window, and S is a constant.
6. A travel user profile building system, based on the travel user profile building method according to any one of claims 1 to 5, characterized in that: It includes a data storage module, a data analysis module, a computation processing module, and an output module; The data storage module is used to collect and save customer information, and can also store feature tags for attractions. The data analysis module is used to classify customer information, and in conjunction with the data storage module, it enables partitioned storage of customer information of different categories. The calculation and processing module is used to calculate the feature tags of customer information and attractions, and to display attractions that customers are interested in to customers. The output module is used to organize the data and output it uniformly to the client APP used by the customer.
7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the travel user profile construction method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the travel user profile construction method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Multi-source intelligent tourism recommendation method based on fine-grained sentiment analysis
CN116257618A
Augmenting text messages with emotion information
US20170147202A1
