Travel motivation recognition method and apparatus, system, storage medium

Abstract

The application discloses a travel motivation recognition method and device, system and storage medium, comprising the following steps: S1, obtaining a user portrait and text information; S2, performing LLM reconstruction, word segmentation processing and feature statistics based on a motivation dictionary on the text information; S3, generating a text motivation prediction result through a machine learning model and rules; S4, predicting a corresponding motivation based on the user portrait; and S5, fusing and strengthening two types of motivation results according to a preset fusion rule and outputting a final motivation score. According to the technical scheme, the user portrait and text semantics are fused, the motivation dictionary and the machine learning model are combined, and accurate recognition and confidence sorting of multiple travel motivations of the user are realized.

Description

Technical Field

[0001] This invention belongs to the field of big data processing technology, specifically relating to a method, device, system, and storage medium for identifying travel motivations. Background Technology

[0002] With the rapid development of online travel platforms, content communities, and short video platforms, users post a large amount of travel-related hashtags, topic descriptions, and text content of varying lengths before and after their trips. These texts implicitly reveal users' travel motivations, such as: relaxation motivations (e.g., "want to lie down and rest," "vacation to unwind"); social motivations (e.g., "playing with friends," "party," "girls' trip"); motivations to connect with nature / sightseeing (e.g., "seeing the sea," "seeing snow-capped mountains," "seeing cherry blossoms"); family / parent-child motivations (e.g., "traveling with kids," "family trip"); and higher-level motivations such as self-growth and self-actualization.

[0003] In existing technologies, the user profiling and content understanding methods commonly used by platforms mainly include: 1. Recommendation models based on behavioral data For example, collaborative filtering, matrix factorization, or deep learning recommendation models can be used to mine preferences from user clicks, favorites, and orders. These methods often only learn "which attractions / content they like" without explicitly extracting "why they like it (what is their motivation)," making them difficult to interpret.

[0004] 2. Content understanding based on topic modeling or simple text classification Some systems use LDA topic models, keyword rules, or ordinary text classifiers to divide content into coarse-grained tags such as "seaside tour," "city tour," and "family tour." However, this approach has several shortcomings: First, it usually only outputs a single main tag, which cannot express the coexistence of multiple travel motivations. Second, it rarely combines user profile information (age, gender, travel type, etc.) for joint inference. Finally, it lacks a systematic model of motivation concepts, focusing more on "scenes / categories" rather than "psychological motivations."

[0005] 3. Simple tagging based solely on rules or dictionaries Another common approach is to manually maintain word lists such as "parent-child related words" and "couple related words," and directly tag words based on keyword matching. This method has the following drawbacks: First, the dictionary coverage is limited and it is difficult to adapt to popular online slang and colloquial expressions; second, it cannot provide confidence or probability, only "yes / no"; and finally, it is not easy to combine with user profiles and has insufficient generalization ability.

[0006] In addition, the following issues were discovered during the analysis of travel platform reviews and user data: Relying solely on text content makes one susceptible to expressive noise, especially with short texts; Relying solely on user profiles ignores the specific topic and description at hand; Existing methods generally cannot uniformly output the Top K travel motivations and their corresponding confidence levels, which is not conducive to subsequent personalized recommendation strategies and empirical research. Summary of the Invention

[0007] To address the problems existing in the prior art, the present invention provides a method, apparatus, system, and storage medium for identifying travel motivations, solving the following problems: 1. Relies solely on user behavior or crude labels, lacking explicit identification of travel motivations; 2. The text motivation recognition failed to fully integrate user profile information, resulting in insufficient accuracy and stability; 3. It does not support TopK confidence ranking of multiple travel motivations for users, resulting in insufficient output information; 4. While solutions based on simple keyword rules are highly interpretable, they are poor in robustness, scalability, and maintainability.

[0008] To achieve the above objectives, the present invention provides the following solution: A method for identifying travel motivations includes: Step S1: Obtain user profiles and text information; Step S2: Perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary on the text information; Step S3: Enhance the generation of text motivation prediction results through machine learning models and rules; Step S4: Based on user profiles, predict corresponding motivations; Step S5: According to the preset fusion rules, the two types of motivation results are fused and enhanced, and the final motivation score is output.

[0009] Preferably, in step S5, the text motivation prediction result text_topk and its probability, the user profile motivation prediction result profile_topk and its score are received, and the final TopK travel motivation list and its confidence level are obtained according to a set of preset reinforcement rules and dynamic fusion logic.

[0010] Preferably, in step S3, the machine learning model is one of RandomForest, Gradient Boosting Tree, or XGBoost.

[0011] The present invention also provides a travel motivation identification device, comprising: The first processing module is used to obtain user profiles and text information; The second processing module is used to perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary for text information. The third processing module is used to enhance the generation of text motivation prediction results through machine learning models and rules; The fourth processing module is used to predict corresponding motivations based on user profiles; The fifth processing module is used to fuse and enhance the two types of motivational results according to preset fusion rules, and output the final motivational score.

[0012] Preferably, the fifth processing module receives the text motivation prediction result text_topk and its probability, the user profile motivation prediction result profile_topk and its score, and obtains the final TopK travel motivation list and its confidence level according to a set of preset reinforcement rules and dynamic fusion logic.

[0013] As a preferred choice, the machine learning model is one of the following: Random Forest, Gradient Boosting Tree, or XGBoost.

[0014] The present invention also provides a travel motivation identification system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a travel motivation identification method when executed by the processor.

[0015] The present invention also provides a storage medium storing a computer program, which executes a travel motivation identification method when running.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention integrates user profile features with topic text semantic features, and introduces a motivation keyword dictionary and machine learning model to output the TopK travel motivations of users in specific topic scenarios and their corresponding confidence scores, thereby improving the accuracy, interpretability and practicality of travel motivation identification. It can be widely applied to online travel platforms, content recommendation systems and tourism research and analysis scenarios. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are 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.

[0018] Figure 1 This is a flowchart of the travel motivation identification method according to an embodiment of the present invention. Detailed Implementation

[0019] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1 like Figure 1 As shown, this invention provides a method for identifying travel motivations. In an input scenario containing user profile information and topic text information, it constructs an overall technical solution combining a motivation dictionary, text feature extraction, a machine learning model, profile rules / models, and fusion and reinforcement rules. The method outputs the user's Top K travel motivations and their confidence levels in that scenario, including: Step S1: Obtain user profiles and text information; Step S2: Perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary on the text information; Step S3: Enhance the generation of text motivation prediction results through machine learning models and rules; Step S4: Based on user profiles, predict corresponding motivations; Step S5: According to the preset fusion rules, the two types of motivation results are fused and enhanced, and the final motivation score is output.

[0022] As one embodiment of the present invention, step S1 includes: Step 11: Receive a request message containing user_profile, topic, topic_desc, and topk; Step 12: Verify whether the fields are complete and whether the data types conform to the preset specifications (e.g., whether topk is an integer and within the range of [1, 10]). If they do not conform to the rules, return an error message. Step 13: Encapsulate the validated request data into an internal unified data structure RequestContext, which includes: context.user_profile, context.topic_list, context.topic_desc, and context.topk.

[0023] Furthermore, in step 11, user_profile: User profile information, which may include: gender: sex; age_group: age group (e.g., "18-25", "26-35", etc.); travel_type: Travel type (e.g., "family" for families with children, "couple" for couples, "friends" for friends traveling together, "solo" for solo travel, etc.). Other profile fields such as city_level and income_level.

[0024] topic: A list of topic tags, such as ["Beach Vacation", "Family Trip"].

[0025] topic_desc: A text description related to this topic, such as "I want to take my child to the beach to play in the sand, see the sea view, and relax a bit."

[0026] topk: The expected number of travel motivations to be output, for example, 3.

[0027] In one embodiment of the present invention, step S2 involves receiving a RequestContext, outputting a text feature vector X_text, and connecting it to the motivation dictionary management module to read motivation dictionary information. Specifically, this includes: Step 21: Topic Text Construction Concatenate multiple tags in topic_list into a string topic_str using semicolons or other preset delimiters; Concatenate topic_str and topic_desc to form a unified topic input text T, for example: "[Seaside Vacation; Family Trip] I want to take my child to the beach to play in the sand, see the sea view, and relax."

[0028] Step 22: Text Preprocessing Use existing Chinese word segmentation tools to segment text T; Use a stop word list to remove meaningless words such as function words and auxiliary words; Retain nouns, verbs, adjectives, and other vocabulary related to travel motivation; Optionally, lowercase conversion, number normalization, and other processing can be performed to generate preprocessed word sequence tokens.

[0029] Step 23: Feature statistics based on the motivation dictionary The motivation dictionary, stored in a database or configuration file, represents the correspondence between various travel motivations and keywords, and includes at least: motivation_id: A unique identifier for the motivation category, such as "RELAX" (relaxation), "SOCIAL" (social motivation), "FAMILY" (parent-child / family motivation), "NATURE" (closeness to nature), "GROWTH" (personal growth), etc. motivation_name: Chinese name; core_keywords: A list of keywords defining motivation, such as relaxation motivation including "lying down", "emptying one's mind", "daydreaming", "vacation", etc. extended_keywords: Expand the target word list to cover colloquial and popular internet slang expressions, such as "lying flat", "emptying oneself", "playing in a Buddhist style", and "electrifying". weight: Keyword weight, used to distinguish between strong and weak indicator words.

[0030] The dictionary is structured so that each keyword is a key, and each key is further divided into verbs, nouns, and adjectives, with each word having a weight. Example: "Yearning for Freedom 1": { "nouns": { "Experience": 0.503 }, "verbs": { "Discovery": 0.5641, "Perception": 0.6395 "Appreciation": 0.507, "Enjoy": 0.4997, "Immersive": 0.8222, "Roaming": 0.8222 }, "adjectives": { "Open": 0.8222, "Tranquility": 0.8222, "Spacious": 0.8222, "Transparency": 0.8222 } }

[0031] Furthermore, the motivation dictionary management module is invoked to load the core_keywords and extended_keywords for each motivation category; For each type of motivation i, the following characteristics are statistically analyzed: count_core_i: The number of times the core keyword appears; count_ext_i: Number of occurrences of extended keywords; weighted_count_i: Weighted term frequency obtained by accumulating different keywords according to their preset weights; coverage_i: The percentage of sentences containing motivational keywords out of all sentences (if divided by sentence).

[0032] Calculate the overall text length len_text, the length after word segmentation len_tokens, and the keyword density density_i = (count_core_i + count_ext_i) / len_tokens, etc.

[0033] Step 24: Feature Vector Construction and Standardization The features of each motivation dimension mentioned above are concatenated with the global features to form the initial feature vector X_raw; Call a pre-trained feature scaler (e.g., StandardScaler) to standardize X_raw, and obtain the text motivation feature vector X_text.

[0034] In one embodiment of the present invention, step S3 involves receiving a text feature vector X_text, calling a pre-trained machine learning model, and outputting a first motivational score vector P_text. Specifically, this includes: Step 31, Model Loading Load the text motivation prediction model Model_text, which can be a machine learning model such as RandomForest, Gradient Boosting Tree, or XGBoost; at the same time, load the feature order information bound to the model to ensure that each dimension in X_text is consistent with that during training.

[0035] Step 32, Prediction Process Input X_text into Model_text; Output the score or probability of each travel motivation category, for example: P_text = (p_text,RELAX, p_text,SOCIAL, p_text,FAMILY, p_text,NATURE, p_text,GROWTH); Store P_text in RequestContext.

[0036] In one embodiment of the present invention, step S4 involves reading the user_profile and connecting it to the motivation fusion and rule reinforcement module to output a second motivation score vector P_profile. Specifically, this includes: Step 41: Image Feature Encoding Standardize the encoding of fields such as gender, age_group, and travel_type. For example, gender can be encoded as 0 / 1; age group can be mapped to a range index; and travel type can be mapped to a multi-dimensional vector through one-hot encoding.

[0037] Step 42, Rule Mapping: By constructing a "profile-motivation mapping table" and inputting travel type, age, and gender, a motivation score can be obtained.

[0038] The table is used to initialize scores for each motivation category, and then normalized to obtain P_profile_rule.

[0039] In one embodiment of the present invention, step S5 involves receiving the text motivation prediction result `text_topk` and its probability, and the user profile motivation prediction result `profile_topk` and its score. Based on a set of preset reinforcement rules and dynamic fusion logic, the final TopK travel motivation list and its confidence level are obtained. Specifically, this includes: Step 51: Strengthen the triggering rules (prioritize text results) The rewritten_query corresponding to the predicted text motivation is subjected to keyword matching. Based on the combination of keyword categories and travel_type, it is determined whether to directly use the predicted text motivation as the final output. For example, when the text contains keywords strongly related to "family companionship" (such as words configured in MOTIVE_TRIGGERS["family companionship 3"]), and the travel type is marked as "solo travel", it is determined that the user's actual scenario is more in line with the need for family companionship, and the top three motivations predicted by the text are directly used as the final result.

[0040] Step 52, Weighted Fusion When the reinforcement rule is not triggered, the text motivation prediction result (text_topk) and the user profile motivation prediction result (profile_topk) are obtained respectively. The overlap set of the two is calculated, and the fusion strategy is dynamically determined according to the degree of overlap: when the number of overlapping motivations is greater than or equal to K (i.e., completely consistent or highly consistent), it is considered that the text and profile judgments of user motivation are highly consistent, and the top K motivations in the overlap set are directly taken as the final result; when the number of overlapping motivations is greater than or equal to K / 2, it is considered that the two directions are generally consistent, but the order may be slightly different. At this time, the text motivation prediction result is dominant, and the top K motivations identified by the text motivation are directly used as the final result; when the number of overlapping motivations is less than K / 2, it is considered that the text result and the profile result are significantly different, and score-driven fusion completion is required.

[0041] Step 53: Score-driven fusion completion When there is little overlap, perform the following steps to complete the fusion: 1) Use the overlapping part of the text results and the portrait results as the initial motivation set; 2) For the remaining motivations in the text results other than the overlapping part, sort them from high to low according to the probability score text_topk_probs, and add the high-scoring motivations to the set in order until the set size reaches K or the text motivations are exhausted; 3) If the set size is still less than K at this time, select the motivations not included in the set from the portrait motivation results, sort them from high to low according to the portrait motivation score, and add them to the set in order until the set contains K motivations or there are no motivations to add.

[0042] Step 54, TopK Output It receives P_final (the list of motivations obtained after score-driven fusion and completion) and topk (the top K results with the highest scores after sorting), and outputs the final list of travel motivations. Specifically, it includes: (1) Sort each travel motivation category in P_final from high to low score to obtain the sorting sequence.

[0043] (2) Read the topk parameter. If topk exceeds the total number of motivation categories, truncate to the total number; if topk is less than 1, use the default value (e.g., 3).

[0044] (3) Select the top K motivation categories with the highest scores and construct a list containing motivation names and scores, for example: { "motivations": [ {"name": "Leisure and Relaxation", "score": 0.89}, {"name": "Parent-child / Family Time", "score": 0.83}, {"name": "Approaching Nature", "score": 0.71} ] } (4) Return the structure to the calling business system for use in recommendation, profile building or research.

[0045] This invention improves the accuracy and stability of travel motivation identification by simultaneously utilizing user profile information and topic text semantics; it supports the coexistence of multiple travel motivations through motivation score vectors and TopK outputs; and it balances interpretability and model performance through a motivation dictionary and rule reinforcement mechanism. The method and system are easily deployed in a service-oriented manner and can be readily integrated into existing tourism platforms and recommendation systems, demonstrating promising application prospects. Using a set of real platform user data, with the results of two human annotators labeling each record with the three most prominent travel motivations as a reference standard, the method of this invention outputs three travel motivation labels for the same record, and the hit rate between the algorithm-output motivations and the human-annotated motivations is statistically analyzed. Experimental results show that, in this embodiment, the Top-3 accuracy rate of the travel motivation identification method of this invention is approximately 86%, which is a significant improvement compared to control methods based solely on text or user profiles, thus further verifying the effectiveness of the technical solution of this invention in practical application scenarios. Example 2 The present invention also provides a travel motivation identification device, comprising: The first processing module is used to obtain user profiles and text information; The second processing module is used to perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary for text information. The third processing module is used to enhance the generation of text motivation prediction results through machine learning models and rules; The fourth processing module is used to predict corresponding motivations based on user profiles; The fifth processing module is used to fuse and enhance the two types of motivational results according to preset fusion rules, and output the final motivational score.

[0046] As one embodiment of the present invention, the fifth processing module receives the text motivation prediction result text_topk and its probability, the user profile motivation prediction result profile_topk and its score, and obtains the final TopK travel motivation list and its confidence level according to a set of preset reinforcement rules and dynamic fusion logic.

[0047] As one embodiment of the present invention, the machine learning model is one of RandomForest, Gradient Boosting Tree, or XGBoost.

[0048] Example 3 The present invention also provides a travel motivation identification system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a travel motivation identification method when executed by the processor.

[0049] Example 4 The present invention also provides a storage medium storing a computer program, which executes a travel motivation identification method when running.

[0050] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for identifying travel motivations, characterized in that, include: Step S1: Obtain user profiles and text information; Step S2: Perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary on the text information; Step S3: Enhance the generation of text motivation prediction results through machine learning models and rules; Step S4: Based on user profiles, predict corresponding motivations; Step S5: According to the preset fusion rules, the two types of motivation results are fused and enhanced, and the final motivation score is output.

2. The travel motivation identification method as described in claim 1, characterized in that, In step S5, the text motivation prediction result text_topk and its probability, the user profile motivation prediction result profile_topk and its score are received, and the final TopK travel motivation list and its confidence level are obtained according to a set of preset reinforcement rules and dynamic fusion logic.

3. The travel motivation identification method as described in claim 2, characterized in that, In step S3, the machine learning model is one of the following: Random Forest, Gradient Boosting Tree, or XGBoost.

4. A travel motivation identification device, characterized in that, include: The first processing module is used to obtain user profiles and text information; The second processing module is used to perform LLM reconstruction, word segmentation, and feature statistics based on the motivation dictionary for text information. The third processing module is used to enhance the generation of text motivation prediction results through machine learning models and rules; The fourth processing module is used to predict corresponding motivations based on user profiles; The fifth processing module is used to fuse and enhance the two types of motivational results according to preset fusion rules, and output the final motivational score.

5. The travel motivation identification device as described in claim 4, characterized in that, The fifth processing module receives the text motivation prediction result text_topk and its probability, the user profile motivation prediction result profile_topk and its score, and obtains the final TopK travel motivation list and its confidence level according to a set of preset reinforcement rules and dynamic fusion logic.

6. The travel motivation identification method as described in claim 5, characterized in that, The machine learning model is one of the following: Random Forest, Gradient Boosting Tree, or XGBoost.

7. A travel motivation identification system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program performing the travel motivation identification method as described in any one of claims 1-3 when executed by the processor.

8. A storage medium, characterized in that, The storage medium stores a computer program that, when executed, performs the travel motivation identification method as described in any one of claims 1-3.

Images