An AI-based travel note automatic generation method and system

Abstract

The application relates to the technical field of intelligent text generation, and discloses a travel note automatic generation method and system based on AI, which comprises the following steps: obtaining an original data set during a travel; clustering the original data to obtain a clustered data group; calculating a time sequence dependency relationship based on the clustered data group, and constructing a time sequence correlation graph; generating a time sequence paragraph sequence according to the correlation graph; calculating the logical fluency among semantic elements and sorting to form a semantic arrangement sequence; performing connection optimization and framework construction on the semantic sequence to obtain a coherent semantic framework; combining the clustered data group to arrange a draft structure and generate a complete note draft; and finally adjusting emotional elements according to the draft to generate travel note content. The application can enhance the semantic coherence and emotional presentation effect of note content.

Description

Technical Field

[0001] This invention relates to the field of intelligent text generation technology, and in particular to an AI-based method and system for automatically generating travel notes. Background Technology

[0002] With the widespread adoption of smart devices and the increasing demand for content recording, travel notes have become an important way for users to express information on social media platforms, personal diaries, and travel services. Especially with the development of mobile internet and intelligent algorithms, users expect to use technology to automatically complete the process of organizing travel records and generating notes, not only saving the time cost of manual editing but also improving the narrative coherence and emotional expression of the content.

[0003] Existing methods for automatically generating travel notes divide and sort activity segments based on timestamps and GPS tracks. This method first obtains the user's location changes and time information during the trip from the mobile device and constructs a travel route sequence accordingly. Then, it binds user-uploaded photos, voice recordings, or text notes to corresponding route nodes, generating a rough tour order based on chronological relationships. However, while existing technologies can achieve basic alignment and text generation based on time and location, their core flaw lies in a lack of contextual understanding of the travel content. They fail to uncover and express the highlights and emotional shifts of the user's journey, resulting in notes that lack narrative structure and fail to evoke emotional resonance.

[0004] Existing technologies suffer from a lack of semantic understanding and emotional expression. Summary of the Invention

[0005] This invention provides an AI-based method and system for automatically generating travel notes, in order to enhance the semantic coherence and emotional presentation of the note content.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides an AI-based method for automatically generating travel notes, comprising:

[0007] Obtain the raw data set during the trip;

[0008] Based on the original dataset, clustering is performed to obtain clustered data groups;

[0009] Based on the clustered data groups, time-series dependencies are calculated, and corresponding time-series association graphs are constructed.

[0010] Based on the aforementioned temporal correlation diagram, a segment sequence is generated to obtain a temporal segment sequence;

[0011] Based on the temporal segment sequence, the logical fluency between semantic elements is calculated, and the elements are sorted according to the logical fluency to obtain a semantic arrangement sequence.

[0012] Based on the semantic arrangement sequence, semantic connection optimization and framework construction are performed to obtain a coherent semantic framework;

[0013] Based on the coherent semantic framework and combined with the clustered data group, the draft structure is organized to generate a complete note draft;

[0014] Based on the complete draft notes, the emotional elements were adjusted to obtain the final travel notes content.

[0015] Preferably, the step of clustering and grouping the original dataset to obtain clustered data groups includes:

[0016] The original data set is sorted by timestamp to obtain an ordered original data sequence;

[0017] In the original data sequence, any two data points are selected as a data point pair, and the spatial distance and time interval of the data point pair are calculated.

[0018] Based on the spatial distance and time interval, the data point pairs are clustered to obtain clustered data groups.

[0019] Preferably, the step of calculating temporal dependencies based on the clustered data groups and constructing corresponding temporal association graphs includes:

[0020] Extract the temporal features of each data point in the clustered data group to obtain a temporal feature data set;

[0021] Based on the aforementioned time-series feature data set, correlation calculations and correlation matrix construction are performed to obtain the correlation strength matrix;

[0022] Based on the association strength matrix, highly associated nodes are identified and an association graph is constructed to obtain a temporal association graph.

[0023] Preferably, the step of generating a segment sequence based on the temporal correlation graph to obtain a temporal segment sequence includes:

[0024] For each node in the time-series correlation graph, correlation filtering is performed to obtain a set of nodes to be merged;

[0025] Determine the continuity of the connection paths between nodes in the set of nodes to be merged. If the paths are continuous, mark each node in the path as a continuous paragraph node belonging to the same path and aggregate them into a continuous paragraph set.

[0026] For nodes in the set of nodes to be merged that are not included in a continuous paragraph, calculate the correlation strength with neighboring nodes. If the correlation strength is lower than a preset strength threshold, mark the corresponding node as an independent node that does not belong to any continuous paragraph and aggregate them into an independent fragment set.

[0027] The continuous paragraph set and the independent segment set are arranged in ascending order by timestamp to generate a time sequence of paragraphs.

[0028] Preferably, the step of calculating the logical fluency between semantic elements based on the temporal segment sequence, and sorting them according to the logical fluency to obtain a semantic arrangement sequence, includes:

[0029] The time sequence segment is mapped to the original log data, and natural language text descriptions are extracted.

[0030] Based on the natural language text description, word segmentation and part-of-speech tagging are performed to obtain a semantic element set;

[0031] Based on the set of semantic elements, sentiment tendency scores are calculated and classified to obtain a sentiment label set;

[0032] Based on the set of semantic elements and the set of sentiment annotations, time interval calculation and context association evaluation are performed to obtain a set of fluency scores. The fluency scores are then sorted to obtain a semantic arrangement sequence.

[0033] Preferably, the step of performing semantic coherence optimization and framework construction based on the semantic arrangement sequence to obtain a coherent semantic framework includes:

[0034] Extract the semantic elements with action descriptions from the semantic arrangement sequence to obtain the action semantic element set;

[0035] For each semantic element in the action semantic element set, calculate the semantic matching score with the semantic elements before and after it. If the semantic matching score is lower than the preset matching threshold, select the corresponding connecting phrase from the preset connecting phrase library and insert it into the corresponding semantic element and the position of the semantic elements before and after it to construct the adjusted semantic sequence.

[0036] Based on the adjusted semantic sequence, a semantic framework is constructed to obtain a coherent semantic framework.

[0037] Preferably, the step of organizing the draft structure based on the coherent semantic framework and combining the clustered data groups to generate a complete draft note includes:

[0038] Based on the coherent semantic framework and combined with the clustered data group, the semantic paragraphs are initially constructed to form the initial draft of the semantic paragraphs;

[0039] Based on the initial draft of the semantic paragraphs, a structure diagram is constructed to obtain the semantic structure diagram;

[0040] The distribution uniformity among nodes in the semantic structure graph is calculated. If the distribution uniformity is lower than a preset uniformity threshold, the semantic structure graph is adjusted in terms of paragraphs and updated in terms of node connections to obtain an adjusted semantic structure graph.

[0041] The adjusted semantic structure diagram is converted into a draft narrative structure, and semantic content is filled and narrative logic is optimized based on the draft narrative structure to obtain a complete draft note.

[0042] Preferably, the step of adjusting the emotional elements based on the complete draft notes to obtain the final travel notes content includes:

[0043] Sentiment semantic analysis is performed on the complete draft notes to identify and extract semantic fragments containing emotion description features, and to construct a semantic set labeled with emotion intensity.

[0044] Based on the emotional elements in the semantic set, emotional clustering and semantic matching are performed to determine the initial position, and the emotional elements are inserted into the initial position to obtain the complete draft notes after insertion.

[0045] For the inserted complete draft notes, calculate the emotional transition smoothing value. If the emotional transition smoothing value is lower than the preset smoothing threshold, then optimize and adjust the emotional elements of the inserted complete draft notes to obtain the final travel note content.

[0046] Secondly, the present invention provides an AI-based automatic travel note generation system, comprising:

[0047] The data acquisition module is used to acquire the raw data set during the trip;

[0048] The clustering module is used to perform clustering and grouping based on the original data set to obtain clustered data groups;

[0049] The time-series module is used to calculate time-series dependencies based on the clustered data groups and construct the corresponding time-series association graph.

[0050] The paragraph sequence module is used to generate a paragraph sequence based on the time-series correlation diagram to obtain a time-series paragraph sequence.

[0051] The sorting sequence module is used to calculate the logical fluency between semantic elements based on the temporal segment sequence, and sort them in order based on the logical fluency to obtain a semantic arrangement sequence.

[0052] The semantic coherence module is used to optimize semantic connection and construct a framework based on the semantic arrangement sequence to obtain a coherent semantic framework.

[0053] The draft module is used to organize the draft structure based on the coherent semantic framework and the clustered data group, and generate a complete note draft.

[0054] The final module is used to adjust the emotional elements based on the complete draft notes to obtain the final travel notes content.

[0055] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the AI-based automatic travel note generation method described in any one of the above.

[0056] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the AI-based automatic travel note generation method described above.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) By performing clustering and temporal-spatial calculations on the original data set during the trip, this invention can identify the boundaries of activity segments in the travel route and effectively divide different tourism scenarios, thereby providing a clear structural foundation for subsequent semantic organization and emotional arrangement.

[0059] (2) By constructing a temporal association diagram and generating a paragraph sequence, this invention can reconstruct the sequential relationship of travel events, making the chapter structure of the notes fit the real travel process and improving the logical coherence and readability of the overall narrative.

[0060] (3) By performing semantic word segmentation, context fluency calculation and sorting on each piece of content, this invention can effectively adjust the order of sentences, avoid context jumps and narrative breaks, and enhance the fluency of text in terms of time and semantics.

[0061] (4) By inserting and adjusting the emotional description content and calculating the smoothness of the transition, this invention can achieve a natural transition of emotional expression in different stages of travel, enhance the appeal of the content and the effect of personalized expression, and make the final notes more realistic in color and more attractive to read. Attached Figure Description

[0062] Figure 1This is a schematic diagram of the AI-based automatic travel note generation method provided in the first embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the AI-based automatic travel note generation system provided in the second embodiment of the present invention. Detailed Implementation

[0064] 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.

[0065] Reference Figure 1 The first embodiment of the present invention provides an AI-based method for automatically generating travel notes, including the following steps:

[0066] S11, Obtain the raw data set during the trip;

[0067] S12, perform clustering and grouping based on the original data set to obtain clustered data groups;

[0068] S13, perform temporal dependency calculation based on the clustered data group, and construct the corresponding temporal association graph;

[0069] S14, Based on the time-series correlation diagram, a segment sequence is generated to obtain a time-series segment sequence;

[0070] S15, calculate the logical fluency between semantic elements according to the temporal segment sequence, and sort them according to the logical fluency to obtain a semantic arrangement sequence;

[0071] S16. Based on the semantic arrangement sequence, perform semantic connection optimization and framework construction to obtain a coherent semantic framework;

[0072] S17. Based on the coherent semantic framework and combined with the clustered data group, the draft structure is organized to generate a complete note draft.

[0073] S18. Based on the complete draft notes, adjust the emotional elements to obtain the final travel notes content.

[0074] In step S11, the original data set during the trip is obtained.

[0075] It's worth noting that this dataset primarily includes two types of information: location timestamps and activity records. Location timestamps refer to the latitude and longitude coordinates and their corresponding acquisition time continuously read by the system from the positioning components (such as a GPS chip) in the terminal, used to describe the user's specific location and time point on the travel route. Activity records originate from the terminal's built-in accelerometer and gyroscope. By reading the output values ​​of these sensors, the system identifies the user's current movement state, such as walking, stationary, or riding in a vehicle. Taking the accelerometer as an example, its output of three-axis acceleration data (X, Y, and Z axes) per second can reflect the frequency and amplitude of the user's body vibrations. The system first continuously acquires acceleration data within a sliding time window (e.g., 3 seconds) and calculates the root mean square (RMS) value and amplitude of change of acceleration within that window. For example, if the root mean square (RMS) value is within 0.5 m / s² and the change is less than 0.2 m / s², the system classifies it as "stationary"; if the RMS value fluctuates periodically between 0.5 and 3.0 m / s², the system classifies it as "walking"; if the fluctuation frequency is high and the changes are drastic, and combined with GPS data, the location is found to be moving rapidly, then it is classified as "riding". In actual operation, the system automatically initiates a sampling request every set time interval (e.g., every 30 seconds) and reads the current location and status of the device. Taking a user walking in a city park between 8:00 AM and 9:00 AM as an example, the system records the following sets of data points: At 8:00 AM, the latitude and longitude are recorded as 39.9000 and 116.3000, and the activity status is "walking"; at 8:05 AM, the location is recorded as 39.9010 and 116.3012, and the status is still "walking"; at 8:15 AM, the location is recorded as 39.9020 and 116.3030, and the status changes to "stationary," at which point the user may be resting by the lake. These data points are recorded using time tags, thus forming a set of raw data with a clear structure and explicit time sequence.

[0076] During the data collection process, to ensure data integrity and continuity, the system sets up a data buffer. When a user is temporarily in a network-free environment, the collected data is stored in a local cache and automatically uploaded to the server for processing once the network is restored. To eliminate outliers with large errors, the system also sets a location jump threshold. For example, if the distance between two adjacent points is greater than 300 meters and the time interval is less than 1 minute, the system marks them as outliers and automatically discards them or proceeds to a subsequent filtering process. Distance and time interval thresholds can be set according to the actual scenario to identify outliers.

[0077] In addition, the system will access the time and location information of photos taken by the user when authorized. For example, if a user stops to take a photo at a certain scenic spot, the Exif information of the photo file will record the GPS coordinates and timestamp at the time of the photo being taken. The system will include this information in the original data set as supplementary information about the location where the user stopped.

[0078] In step S12, clustering is performed based on the original dataset to obtain clustered data groups, including:

[0079] The original data set is sorted by timestamp to obtain an ordered original data sequence;

[0080] In the original data sequence, any two data points are selected as a data point pair, and the spatial distance and time interval of the data point pair are calculated.

[0081] Based on the spatial distance and time interval, the data point pairs are clustered to obtain clustered data groups.

[0082] It's worth noting that the system sorts the original dataset according to its timestamps, ensuring that all data is arranged in chronological order of occurrence. This sorting is done by comparing the timestamps of each data point, arranging them from smallest to largest. For example, if three data entries are recorded at 08:00, 08:05, and 07:55 respectively, the sorted result will be an ordered sequence of 07:55, 08:00, and 08:05.

[0083] Then, two data points are randomly selected from the ordered data sequence as a pair, and their spatial distance and time interval are calculated one by one. The spatial distance is calculated using latitude and longitude coordinates as input, processed using the Euclidean distance formula. Specifically, the system first converts the latitude and longitude information of the two data points into planar coordinates in meters, and then obtains the straight-line distance between the two points by summing the square differences and taking the square root. For example, if the coordinates of data point A are 39.90, 116.30 and the coordinates of point B are 39.91, 116.31, the converted horizontal distance is approximately 100 meters, and the vertical distance is also approximately 100 meters, so the calculated result is approximately 141.4 meters. Simultaneously, the system calculates the time interval between the two data points by comparing their timestamps, in minutes or seconds. For example, if the timestamps of the two points are 08:00 and 08:07, the time interval is 7 minutes.

[0084] After obtaining the spatial distances and time intervals of all data point pairs, the system analyzes and judges these values. If the spatial distance between a pair of data points is less than 200 meters and the time interval is less than 10 minutes, clustering is performed, specifically using the K-means method. An initial cluster size K is set, an integer between 3 and 6, evaluated based on the geographical span of the travel data coverage area and the number of stops in the user's itinerary. In practical applications, the system executes a complete clustering process once for each of the four possible K values: 3, 4, 5, and 6, and calculates the intra-cluster average distance and inter-cluster average distance for each clustering result. The intra-cluster distance is obtained by calculating the average distance from all points within each group to the center of that group; the inter-cluster distance is obtained by calculating the average distance between all pairwise group centers. Taking 120 GPS data points collected during a trip as an example, when K=3, the average distance within a cluster is 225 meters, and the average distance between clusters is 670 meters; when K=4, the average distance within a cluster is 140 meters, and the average distance between clusters is 710 meters; when K=5, although the distance between clusters increases to 740 meters, the decrease in the distance within a cluster is small, only 130 meters, and there are cases where a cluster contains only 2 points, resulting in imbalanced clustering. Therefore, K=4 was ultimately chosen as the number of clusters.

[0085] After determining K=4, the system randomly selects four points from the original data as initial centers, for example, point A (latitude 39.900, longitude 116.300), point B (39.920, 116.340), point C (39.910, 116.280), and point D (39.905, 116.320). Subsequently, the system calculates the spatial distance between each of the remaining 116 data points and the four center points. The calculation method involves converting the latitude and longitude coordinates to metric plane coordinates and then using the Euclidean algorithm. For example, point E (39.904, 116.318) is approximately 90 meters from point D and more than 200 meters from the other three center points; therefore, point E is assigned to the group centered on point D.

[0086] After the initial allocation of all data points, the system calculates the average latitude and longitude coordinates of all points within each group to obtain new cluster centers. Taking group D as an example, if there are 30 data points in the group with an average latitude of 39.9062 and an average longitude of 116.3195, then the new center point for group D is set as (39.9062, 116.3195). The system then repeats the aforementioned allocation process with these four new center points until the cluster center changes in two consecutive rounds are less than a set threshold (e.g., 1 meter) or the maximum number of iterations is reached (e.g., 20 rounds), finally outputting a stable clustering result.

[0087] Taking a user's entire day's itinerary in the city as an example, the data was ultimately divided into four clustered data groups: "Morning hotel departure area", "Morning park visit area", "Noon museum visit area", and "Evening shopping mall stay area".

[0088] In step S13, based on the clustered data groups, temporal dependencies are calculated, and corresponding temporal association graphs are constructed, including:

[0089] Extract the temporal features of each data point in the clustered data group to obtain a temporal feature data set;

[0090] Based on the aforementioned time-series feature data set, correlation calculations and correlation matrix construction are performed to obtain the correlation strength matrix;

[0091] Based on the association strength matrix, highly associated nodes are identified and an association graph is constructed to obtain a temporal association graph.

[0092] It's worth noting that the system extracts the timestamps, location change information, and activity type records of data points from each cluster, and then arranges this data in chronological order. For the arranged data sequence, the system uses a sliding analysis approach to sequentially extract each pair of data points that are adjacent in time or whose time intervals are within a preset range, recording their time interval, displacement distance, and speed change value. For example, if a user is visiting a city park, and two points are at 08:00 and 08:05 respectively, with a location difference of 120 meters, then the speed is 24 meters per minute; this pair of data is the valid time pair.

[0093] Next, the system iterates through all data point pairs that meet the criteria within each cluster, and calculates the correlation coefficient for these pairs. This calculation uses the Pearson correlation coefficient to measure the consistency of temporal behavior between two data points. Specifically, the time intervals and speed changes are first standardized to ensure they have the same dimensions. Then, the covariance of the two data points is calculated, and finally, the covariance is divided by the product of the two standard deviations to obtain the Pearson coefficient. This coefficient ranges from -1 to 1, with values ​​closer to 1 indicating more similar temporal behavior between the two points and a stronger continuity in their behavior. For example, if a pair of data points has a Pearson coefficient of 0.86 calculated based on the standardized features, it means that the user's actions at these two points exhibit a stable movement trend, belonging to the same continuous activity sequence.

[0094] All calculated correlation coefficient values ​​will be stored in an array in the form of a two-dimensional table, which is the correlation matrix. Each row and column of the matrix corresponds to a specific data point, and the value in the i-th row and j-th column of the matrix represents the temporal correlation strength between the i-th point and the j-th point.

[0095] The system then determines the correlation strength value of each pair of data points in the correlation strength matrix to decide whether to establish a connection relationship in the graph structure between them. If the coefficient value between two data points is higher than a preset connection threshold, the system establishes a directed edge between these two points, using the coefficient value as the edge weight to represent the chronological order and strength of the edge in the time series. The preset connection threshold is set to 0.7, which, according to statistics from actual travel data samples, indicates that point pairs with a Pearson coefficient greater than 0.7 exhibit continuous and stable movement behavior.

[0096] When constructing the graph structure, the system uses an adjacency matrix to record edges. This involves performing a threshold filtering operation on the original correlation matrix, retaining only elements with values ​​greater than 0.7 and marking them as valid connections in the graph. Each non-zero element in the matrix represents a directed edge in the graph, with its starting point at the corresponding row index node and its ending point at the column index node. The weight of the edge is the corresponding correlation coefficient value. After edge filtering, the system integrates all valid nodes and edges to form a complete directed graph structure, which is then output as a time-series correlation graph.

[0097] In step S14, based on the time-series correlation diagram, a segment sequence is generated to obtain a time-series segment sequence, including:

[0098] For each node in the time-series correlation graph, correlation filtering is performed to obtain a set of nodes to be merged;

[0099] Determine the continuity of the connection paths between nodes in the set of nodes to be merged. If the paths are continuous, mark each node in the path as a continuous paragraph node belonging to the same path and aggregate them into a continuous paragraph set.

[0100] For nodes in the set of nodes to be merged that are not included in a continuous paragraph, calculate the correlation strength between them and their neighboring nodes. If the correlation strength is lower than a preset strength threshold, mark the node as an independent node that does not belong to any continuous paragraph and aggregate it into an independent fragment set.

[0101] The continuous paragraph set and the independent segment set are arranged in ascending order by timestamp to generate a time sequence of paragraphs.

[0102] It is worth noting that, after initial screening of all node pairs in the time-series correlation graph, the system extracts all node connection pairs with a correlation strength value greater than 0.85 as strong correlation candidates and includes these nodes in the "set of nodes to be merged". This threshold is selected to ensure that only closely related data points are processed during the subsequent paragraph merging process, avoiding a decrease in paragraph segmentation accuracy due to weak correlation connections.

[0103] Then, starting from any node in the set of nodes to be merged, the system checks whether it can reach other nodes in the set through the established graph edges. This process involves gradually following each edge with sufficient connection strength, jumping from one node to the next, until all target nodes in the set have been visited. If all nodes can be connected by edges in the graph, the path connectivity is considered valid. Next, the system reads the timestamps of each node on these paths and compares them one by one in the order of access to confirm whether they are arranged in chronological order from earliest to latest. If the timestamp of a node in the path is earlier than that of its predecessor, it indicates a time regression, and the system considers this path to not meet the timing requirements, and the entire node group is also judged as discontinuous. Only when nodes can be successfully connected and all node timestamps are strictly increasing are these nodes marked as a continuous segment and aggregated into a continuous segment set.

[0104] For each node not included in a continuous segment, its timestamp and spatial location in the original data are extracted. Then, its two closest nodes in time are found, and the correlation strength between the node and these two neighbors is calculated. The correlation strength is calculated using the Pearson correlation coefficient, which involves standardizing the timestamps and location changes of the three nodes to obtain their standard deviation and covariance, and then calculating the Pearson coefficient accordingly. If the correlation strength between the node and any of its immediate neighbors is lower than a set independence threshold (e.g., less than 0.7), the system marks the node as an independent segment and does not include it in any continuous segment. These nodes are collectively collected into a set of independent segments to represent phenomena such as user stops, derailments, or short-term random movements during travel.

[0105] Finally, the identified set of consecutive paragraphs and the set of independent segments are integrated and sorted in ascending order according to the start timestamp of each paragraph or segment, thereby generating a complete and chronologically consistent reconstructed sequence of time-series paragraphs.

[0106] To facilitate understanding of the present invention, some preferred embodiments of the present invention will be described in further detail below.

[0107] In step S15, based on the temporal segment sequence, the logical fluency between semantic elements is calculated, and the elements are ordered according to the logical fluency to obtain a semantic arrangement sequence, including:

[0108] The time sequence segment is mapped to the original log data, and natural language text descriptions are extracted.

[0109] Based on the natural language text description, word segmentation and part-of-speech tagging are performed to obtain a semantic element set;

[0110] Based on the set of semantic elements, sentiment tendency scores are calculated and classified to obtain a sentiment label set;

[0111] Based on the set of semantic elements and the set of sentiment annotations, time interval calculation and context association evaluation are performed to obtain a set of fluency scores. The fluency scores are then sorted to obtain a semantic arrangement sequence.

[0112] It's worth noting that, based on the chronological sequence of paragraphs, the system extracts the corresponding original natural language log content for each paragraph. For the text content in each paragraph, the system uses a word segmentation tool to break down the sentence structure word by word and labels each word with its part of speech. Then, the system filters out keywords containing descriptions of actions and events based on their parts of speech, such as verbs and nouns, as components of the semantic element set. For example, if an original description is "The user slowly walked through the ancient street and took photos," the system will extract "walked through the ancient street" and "took photos" as valid semantic elements, while also recording their position and start and end times within the paragraph.

[0113] The system then iterates through each text element in the semantic element set and uses the pre-built sentiment tendency scoring table to assess it. Each sentiment phrase is assigned a score range associated with three sentiment categories: positive, neutral, and negative. For example, the phrase "taking a photo" has a co-occurrence rate with positive words exceeding 80%, therefore it scores +0.82 and is classified as positive sentiment; the phrase "long queue time" has a high co-occurrence rate with negative words, scoring -0.76 and classified as negative sentiment; while "entering the park" scores +0.05 and is judged as neutral.

[0114] This sentiment rating scale consists of 10,000 Chinese phrases and sentences manually labeled as "positive," "neutral," or "negative," such as "happy," "traffic congestion," and "long queue time." Each sample is assigned a sentiment category label during the training phase, and its frequency of occurrence in similar texts and its co-occurrence with typical sentiment words are calculated through statistical analysis. For example, phrases like "take a photo" appear frequently in the "positive" category and co-occur significantly more often with positive words like "like" and "happy" than in other categories, thus receiving a higher positive sentiment score. The system constructs a co-occurrence matrix and calculates the association degree between each word or phrase and three sentiment center phrases (such as "happy," "normal," and "annoying"), outputting a score within the range [-1, 1].

[0115] Specifically, if a phrase has the highest co-occurrence probability in the "positive" category, the system sets its sentiment score to be greater than or equal to 0.6; if it has a higher co-occurrence frequency in the "negative" category, it is assigned a value less than or equal to -0.6; if the phrase's correlation with the three types of central phrases is close, the system sets it to neutral sentiment, with a score range between -0.4 and 0.4. Taking the phrase "appreciate ancient architecture" as an example, its frequency in positive samples is 82%, in negative samples it is 3%, and in neutral samples it is 15%, so the system assigns it a sentiment score of +0.82, classifying it as positive sentiment; similarly, "waiting too long" has a frequency of 76% in negative corpus, so the system assigns it a sentiment score of -0.76, classifying it as negative sentiment; if a sentence like "enter the park" does not carry obvious emotional words, its frequency of occurrence in each category is balanced, and its final score is approximately 0.05, classifying it as neutral sentiment. Afterwards, the scores of all semantic elements are uniformly recorded in the sentiment annotation set.

[0116] To calculate the fluency score, each reconstructed paragraph is first traversed to extract its semantic elements. The time interval between adjacent elements, in minutes, is then calculated based on their original timestamps for subsequent association strength calculation. Based on this, the system inputs the text content of each semantic element into a pre-loaded word vector library for encoding, resulting in a corresponding dense vector representation. The word vector library is derived from the joint training results of Chinese Wikipedia and publicly available travel log corpora, employing a 300-dimensional embedding vector representation method. For example, "entering the ancient street" is mapped to a 300-dimensional real-valued vector in the word vector.

[0117] Subsequently, cosine similarity is calculated for the word vectors of two adjacent semantic elements. The calculation process includes: first, calculating the dot product of the two vectors, then calculating the magnitude of the two vectors, and finally dividing the dot product by the product of the magnitudes, i.e., similarity = A·B / (‖A‖ × ‖B‖). For example, the similarity result between the vector of element A "entering the ancient street" and the vector of element B "taking a photo" is 0.91, indicating that they are semantically related, and the system sets this as the initial score for logical fluency.

[0118] To further improve the semantic ranking accuracy, the system incorporates sentiment annotation to correct the fluency scores between adjacent elements. The comparison method involves determining whether the sentiment category labels calculated in previous steps are consistent for two elements. If both elements are labeled "positive" or "negative," meaning they have the same sentiment category, a weight compensation of 0.05 is added to the original score; if the sentiments are opposite, 0.05 is subtracted. This compensation value is based on the fitting analysis results of 5000 manually annotated corpora, and the ±0.05 adjustment effectively improves the consistency of ranking within paragraphs and user acceptance.

[0119] After obtaining these scores, all semantic elements within the paragraph are reordered. The elements are sorted sequentially according to their fluency scores to obtain a semantic arrangement sequence.

[0120] In step S16, based on the semantic arrangement sequence, semantic coherence optimization and framework construction are performed to obtain a coherent semantic framework, including:

[0121] Extract the semantic elements with action descriptions from the semantic arrangement sequence to obtain the action semantic element set;

[0122] For each semantic element in the action semantic element set, calculate the semantic matching score with the semantic elements before and after it. If the semantic matching score is lower than the preset matching threshold, select the corresponding connecting phrase from the preset connecting phrase library and insert it into the corresponding semantic element and the position of the semantic elements before and after it to construct the adjusted semantic sequence.

[0123] Based on the adjusted semantic sequence, a semantic framework is constructed to obtain a coherent semantic framework.

[0124] It's worth noting that all semantic elements containing action descriptions are selected from the semantic sequence. Action-related semantic elements include verb phrases describing a person's behavior or the progression of an event, such as "depart for the park," "pass through the alley," and "arrive at the viewing platform." This operation extracts sentences containing verbs or verb phrases by traversing each element in the semantic sequence and combining the results of word segmentation and part-of-speech tagging. These are then compiled into a set of action semantic elements. Each element includes text content and a sequence position index, facilitating subsequent analysis and processing.

[0125] Subsequently, the system sequentially selects each element from the set of semantic elements for that action and calculates the semantic matching degree with its preceding and following semantic elements. The matching degree is calculated as follows: first, the text of each semantic element is converted into a 300-dimensional real vector using the loaded Word2Vec word vector tool; then, the cosine similarity between the current element and the vector of its preceding element is calculated. The formula is: the dot product of the two vectors divided by their modulus product; the result is between 0 and 1, with a value closer to 1 indicating greater semantic similarity. The system sets a minimum reasonable threshold of 0.55 for the matching degree. When the similarity between a transitional element and its context elements is lower than this value, the system considers the semantic connection at that point to be weak, which may lead to a jump in user understanding or a logical break.

[0126] To address the aforementioned issues, the system selects the most suitable connective words or phrases from a pre-defined corpus of connecting phrases to insert between the two elements. This corpus is manually compiled based on tourism text data and covers commonly used phrases such as "following," "next," "soon after," "immediately afterward," and "then." Each phrase is accompanied by an applicable scenario tag, such as "action connection," "location transfer," and "event progression." Before insertion, the system compares the phrase category and usage context based on contextual features, prioritizing phrases that match the semantic intent. The insertion operation involves directly inserting the connecting phrase before the current action semantic element in the semantic sequence. For example, if the original sequence is "The user departs, passes through the alley, and arrives at the viewing platform," and the match degree between "passes through the alley" and "arrives at the viewing platform" is only 0.48, the system adjusts it to "The user departs, passes through the alley, and then arrives at the viewing platform" to improve logical coherence.

[0127] Finally, the system takes the semantic sequence, after phrase supplementation and structural adjustment, as input, reorganizes the semantic order, and generates a semantic connection graph. Each semantic element is regarded as a node in the graph, and the connecting phrases are used as the annotation content of the edges, while also recording transitional semantic relationships, thus forming a complete and coherent semantic framework.

[0128] In step S17, based on the coherent semantic framework and combined with the clustered data group, the draft structure is organized to generate a complete note draft, including:

[0129] Based on the coherent semantic framework and combined with the clustered data group, the semantic paragraphs are initially constructed to form the initial draft of the semantic paragraphs;

[0130] Based on the initial draft of the semantic paragraphs, a structure diagram is constructed to obtain the semantic structure diagram;

[0131] The distribution uniformity among nodes in the semantic structure graph is calculated. If the distribution uniformity is lower than a preset uniformity threshold, the semantic structure graph is adjusted in terms of paragraphs and updated in terms of node connections to obtain an adjusted semantic structure graph.

[0132] The adjusted semantic structure diagram is converted into a draft narrative structure, and semantic content is filled and narrative logic is optimized based on the draft narrative structure to obtain a complete draft note.

[0133] It's worth noting that the system extracts representative semantic elements from each clustered data group based on the order of each node in the semantic framework. Taking a travel log as an example, if the clustering results are divided into three groups: "Preparation for Departure," "Sightseeing," and "End of Activity," the system will select the element with the highest logical position from each group as the starting sentence of the paragraph. For example, it might select "The user packs their luggage in the morning" from the "Preparation for Departure" group and "Entering the park to admire the flowers and plants" from the "Sightseeing" group.

[0134] The system then arranges the extracted semantic elements in sequence and organizes the text into segments, initially forming a set of semantic paragraphs. The system assigns a start time and end time to each segment and marks the connections between semantic elements within the segment. For example, in the segment "The user packs their luggage in the morning and then goes to the station," the system records the start time as 08:00 and the end time as 08:45. The connecting phrase "then" between the internal elements "packing luggage" and "going to the station" is used to maintain logical flow.

[0135] Next, the system constructs a semantic structure graph based on this set of semantic paragraphs, treating each paragraph as a node in the graph. Connection edges are introduced between nodes according to their original chronological order and logical turning points. Each connection edge is assigned a weight, the value of which is determined by the time interval between paragraphs and their semantic similarity. For example, if the time interval between "packing luggage" and "going out to take a car" is 10 minutes and their semantic similarity is 0.82, then the weight of the connection edge would be 0.82.

[0136] Then, in the statistical semantic structure graph, the degree values ​​of each node are statistically distributed, and the mean degree and standard deviation are calculated. The mean degree is calculated by summing the degrees of all nodes and dividing by the number of nodes; the standard deviation is obtained by averaging the squares of the differences between the degree of each node and the mean degree, and then taking the square root. For example, if there are 6 paragraph nodes with connectivity degrees of 2, 3, 4, 6, 2, and 3 respectively, the mean degree is 3.33 and the standard deviation is 1.25.

[0137] The system sets a preset standard deviation threshold of 1.5. When the standard deviation of the graph structure exceeds this threshold and the mean degree deviates from the target value by ±0.5, it is considered that the semantic structure has an over-concentrated or dispersed distribution problem.

[0138] Then, three specific adjustment operations will be performed on the paragraphs in the unbalanced areas to improve the balance and coherence of the overall semantic structure. First, for paragraphs with a high average connectivity, the system will reclassify the semantic elements within that paragraph. This process involves: if an element has a higher match with a subsequent paragraph than the current paragraph, it will be excluded and merged into the subsequent paragraph, forming a new paragraph boundary. This division method ensures that each paragraph retains only semantically close and temporally connected content, avoiding excessive clustering of nodes.

[0139] Secondly, for paragraphs with sparse connections and unnatural transitions, the system will introduce additional transition phrases to enhance the connection between paragraphs. The process includes: selecting suitable phrases from a pre-set connection phrase library, such as "then," "at the same time," and "immediately following," and inserting them between the two paragraphs, preferably at the end of the preceding paragraph or the beginning of the following paragraph. After semantic restructuring, the system will update the connection edges between nodes based on the latest paragraph division and connection relationships. The update process includes: reconstructing the semantic structure graph, traversing paragraph nodes one by one, and determining whether to retain the original connections, delete low-intensity edges, or add new connections based on the new semantic grouping and the insertion of connection phrases. For example, if there was no connection between two paragraphs before, an edge will be added if a new phrase is inserted; otherwise, it will be deleted.

[0140] Finally, the system transcribes the adjusted semantic structure diagram into a draft structure text with chronological order, logical connection, and semantic transition. Then, based on the cluster group characteristics of each paragraph, it performs content supplementation and word order adjustment. For example, it adds a summary sentence to the paragraphs in the "activity ended" group, such as "This tour was fulfilling and enjoyable", forming a complete draft of notes with an introduction, process, and conclusion.

[0141] In step S18, based on the complete draft notes, emotional elements are adjusted to obtain the final travel notes content, including:

[0142] Sentiment semantic analysis is performed on the complete draft notes to identify and extract semantic fragments containing emotion description features, and to construct a semantic set labeled with emotion intensity.

[0143] Based on the emotional elements in the semantic set, emotional clustering and semantic matching are performed to determine the initial position, and the emotional elements are inserted into the initial position to obtain the complete draft notes after insertion.

[0144] For the inserted complete draft notes, calculate the emotional transition smoothing value. If the emotional transition smoothing value is lower than the preset smoothing threshold, then optimize and adjust the emotional elements of the inserted complete draft notes to obtain the final travel note content.

[0145] It's worth noting that the entire text of the complete draft notes was scanned segment by segment to extract semantic fragments with obvious emotional expression. The identification of these fragments relied on a pre-built sentiment lexicon. This lexicon was obtained through manual annotation and statistical analysis from a large amount of travel-related text, containing common sentiment terms such as "excitement," "moved," "bored," and "shocked." Each term is accompanied by a sentiment direction label (positive or negative) and a corresponding sentiment intensity value. The sentiment direction is determined based on the word's meaning; for example, "moved" is marked as positive, and "disappointed" as negative. The sentiment intensity value is calculated based on the average subjective rating associated with the word's appearance in historical travel notes. For example, in 2000 user travelogues, the word "shocked" received an average rating of 9.2 out of 10, and the system assigned it a value of 1.4 after standardization of its sentiment intensity.

[0146] After initial extraction, the system clusters all emotional fragments according to their semantic features. For example, "moved," "shocked," and "excited" are grouped into the climax category, while "tranquil," "comfortable," and "peaceful" are grouped into the background category. Following clustering, semantic matching is performed based on these emotional categories and the semantic types of each paragraph in the note structure. Specifically, the matching operation compares the word vectors of the emotional fragments with the word vectors of the original paragraphs, determining the degree of matching by calculating the cosine similarity between the two. For example, the cosine similarity between the vectors of "excited" and "climbing the mountain" is 0.83, which is greater than the preset threshold of 0.8, so the system includes it in the candidate set. If the similarity between a fragment and any position is less than 0.6, the fragment is marked as unsuitable for insertion and processing is temporarily suspended.

[0147] To determine the logical coherence of the inserted text, the system combines the inserted text into three segments, one before and one after the insertion point, and calculates the smoothness of the transition. The specific calculation process is as follows: word vectors are generated for each of the three segments; the semantic similarity (cosine similarity value 1) between the preceding and following segments is calculated; then, the similarity values ​​2 and 3 between the inserted segment and the preceding and following segments are calculated. If the difference between similarity value 2 and similarity value 1 is too large, for example, exceeding 0.3, the system considers the insertion to have disrupted the naturalness of the transition. The smoothness score is the average of these three similarity values; if it is less than a preset smoothness threshold, such as 0.72, it indicates that the insertion is not smooth enough.

[0148] At this point, the system will modify the intensity of the inserted emotional paragraphs. For example, it might rewrite "shocking" as "deeply impressed," and shift from a main statement to a parallel description to soften the impact. Then, it will re-evaluate the smoothness of the three paragraphs until the score exceeds the smoothness threshold. Finally, through multiple rounds of filtering and adjustments, the system produces a final travel journal entry with a natural emotional flow and clear structure.

[0149] Reference Figure 2The second embodiment of the present invention provides an AI-based automatic travel note generation system, comprising:

[0150] The data acquisition module is used to acquire the raw data set during the trip;

[0151] The clustering module is used to perform clustering and grouping based on the original data set to obtain clustered data groups;

[0152] The time-series module is used to calculate time-series dependencies based on the clustered data groups and construct the corresponding time-series association graph.

[0153] The paragraph sequence module is used to generate a paragraph sequence based on the time-series correlation diagram to obtain a time-series paragraph sequence.

[0154] The sorting sequence module is used to calculate the logical fluency between semantic elements based on the temporal segment sequence, and sort them in order based on the logical fluency to obtain a semantic arrangement sequence.

[0155] The semantic coherence module is used to optimize semantic connection and construct a framework based on the semantic arrangement sequence to obtain a coherent semantic framework.

[0156] The draft module is used to organize the draft structure based on the coherent semantic framework and the clustered data group, and generate a complete note draft.

[0157] The final module is used to adjust the emotional elements based on the complete draft notes to obtain the final travel notes content.

[0158] It should be noted that the AI-based travel note automatic generation system provided in this embodiment of the invention is used to execute all the process steps of the AI-based travel note automatic generation method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0159] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an AI-based automatic travel note generation program. When the processor executes the computer program, it implements the steps in the various AI-based automatic travel note generation method embodiments described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the paragraph sequence module.

[0160] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0161] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0162] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0163] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0164] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0165] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0166] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

Claims

1. An AI-based method for automatically generating travel notes, characterized in that, include: Obtain the raw data set during the trip; Based on the original dataset, clustering is performed to obtain clustered data groups; Based on the clustered data groups, time-series dependencies are calculated, and corresponding time-series association graphs are constructed. Based on the aforementioned temporal correlation diagram, a segment sequence is generated to obtain a temporal segment sequence; Based on the temporal segment sequence, the logical fluency between semantic elements is calculated, and the elements are sorted according to the logical fluency to obtain a semantic arrangement sequence. Based on the semantic arrangement sequence, semantic connection optimization and framework construction are performed to obtain a coherent semantic framework; Based on the coherent semantic framework and combined with the clustered data group, the draft structure is organized to generate a complete note draft; Based on the complete draft notes, the emotional elements were adjusted to obtain the final travel notes content; The step of generating a segment sequence based on the temporal correlation graph to obtain a temporal segment sequence includes: For each node in the time-series correlation graph, correlation filtering is performed to obtain a set of nodes to be merged; Determine the continuity of the connection paths between nodes in the set of nodes to be merged. If the paths are continuous, mark each node in the path as a continuous paragraph node belonging to the same path and aggregate them into a continuous paragraph set. For nodes in the set of nodes to be merged that are not included in a continuous paragraph, calculate the correlation strength with neighboring nodes. If the correlation strength is lower than a preset strength threshold, mark the corresponding node as an independent node that does not belong to any continuous paragraph and aggregate them into an independent fragment set. The continuous paragraph set and the independent segment set are arranged in ascending order by timestamp to generate a time sequence of paragraphs; The step of adjusting emotional elements based on the complete draft notes to obtain the final travel notes content includes: Sentiment semantic analysis is performed on the complete draft notes to identify and extract semantic fragments containing emotion description features, and to construct a semantic set labeled with emotion intensity. Based on the emotional elements in the semantic set, emotional clustering and semantic matching are performed to determine the initial position, and the emotional elements are inserted into the initial position to obtain the complete draft notes after insertion. For the inserted complete draft notes, calculate the emotional transition smoothing value. If the emotional transition smoothing value is lower than the preset smoothing threshold, optimize and adjust the emotional elements of the inserted complete draft notes to obtain the final travel note content. Specifically, the complete draft notes undergo sentiment semantic analysis to identify and extract semantic fragments containing sentiment description features, and a semantic set labeled with sentiment intensity is constructed, including: The entire text of the complete draft notes was scanned segment by segment to extract semantic segments with obvious emotional expression; each word was accompanied by an emotional direction tag and a corresponding emotional intensity value; the emotional intensity value was calculated based on the average subjective evaluation score associated with the word when it appeared in historical travel notes.

2. The AI-based automatic travel note generation method according to claim 1, characterized in that, The step of clustering and grouping the data based on the original dataset to obtain clustered data groups includes: The original data set is sorted by timestamp to obtain an ordered original data sequence; In the original data sequence, any two data points are selected as a data point pair, and the spatial distance and time interval of the data point pair are calculated. Based on the spatial distance and time interval, the data point pairs are clustered to obtain clustered data groups.

3. The AI-based automatic travel note generation method according to claim 1, characterized in that, The step of calculating temporal dependencies based on the clustered data groups and constructing corresponding temporal association graphs includes: Extract the temporal features of each data point in the clustered data group to obtain a temporal feature data set; Based on the aforementioned time-series feature data set, correlation calculations and correlation matrix construction are performed to obtain the correlation strength matrix; Based on the association strength matrix, highly associated nodes are identified and an association graph is constructed to obtain a temporal association graph.

4. The AI-based automatic travel note generation method according to claim 1, characterized in that, The step of calculating the logical fluency between semantic elements based on the temporal segment sequence, and sorting them according to the logical fluency to obtain a semantic arrangement sequence includes: The time sequence segment is mapped to the original log data, and natural language text descriptions are extracted. Based on the natural language text description, word segmentation and part-of-speech tagging are performed to obtain a semantic element set; Based on the set of semantic elements, sentiment tendency scores are calculated and classified to obtain a sentiment label set; Based on the set of semantic elements and the set of sentiment annotations, time interval calculation and context association evaluation are performed to obtain a set of fluency scores. The fluency scores are then sorted to obtain a semantic arrangement sequence.

5. The AI-based automatic travel note generation method according to claim 1, characterized in that, The step of optimizing semantic coherence and constructing a framework based on the semantic arrangement sequence to obtain a coherent semantic framework includes: Extract the semantic elements with action descriptions from the semantic arrangement sequence to obtain the action semantic element set; For each semantic element in the action semantic element set, calculate the semantic matching score with the semantic elements before and after it. If the semantic matching score is lower than the preset matching threshold, select the corresponding connecting phrase from the preset connecting phrase library and insert it into the corresponding semantic element and the position of the semantic elements before and after it to construct the adjusted semantic sequence. Based on the adjusted semantic sequence, a semantic framework is constructed to obtain a coherent semantic framework.

6. The AI-based method for automatically generating travel notes according to claim 1, characterized in that, Based on the coherent semantic framework and combined with the clustered data groups, the draft structure is organized to generate a complete draft note, including: Based on the coherent semantic framework and combined with the clustered data group, the semantic paragraphs are initially constructed to form the initial draft of the semantic paragraphs; Based on the initial draft of the semantic paragraphs, a structure diagram is constructed to obtain the semantic structure diagram; The distribution uniformity among nodes in the semantic structure graph is calculated. If the distribution uniformity is lower than a preset uniformity threshold, the semantic structure graph is adjusted in terms of paragraphs and updated in terms of node connections to obtain an adjusted semantic structure graph. The adjusted semantic structure diagram is converted into a draft narrative structure, and semantic content is filled and narrative logic is optimized based on the draft narrative structure to obtain a complete draft note.

7. An AI-based automatic travel note generation system, used to implement the AI-based automatic travel note generation method as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire the raw data set during the trip; The clustering module is used to perform clustering and grouping based on the original data set to obtain clustered data groups; The time-series module is used to calculate time-series dependencies based on the clustered data groups and construct the corresponding time-series association graph. The paragraph sequence module is used to generate a paragraph sequence based on the time-series correlation diagram to obtain a time-series paragraph sequence; The sorting sequence module is used to calculate the logical fluency between semantic elements based on the temporal segment sequence, and sort them in order based on the logical fluency to obtain a semantic arrangement sequence. The semantic coherence module is used to optimize semantic connections and construct a framework based on the semantic arrangement sequence to obtain a coherent semantic framework. The draft module is used to organize the draft structure based on the coherent semantic framework and the clustered data group, and generate a complete note draft. The final module is used to adjust the emotional elements based on the complete draft notes to obtain the final travel notes content.

Images