A life course data analysis method and system

Abstract

The application discloses a life course data analysis method and system, relates to the technical field of data analysis, and comprises the following steps: obtaining historical user data; performing event clustering on the historical user data to obtain event units; performing feature extraction on the event units to obtain core elements and emotional tendencies; generating semantic anchors of the event units according to the core elements and the emotional tendencies; performing correlation analysis on current user data and the semantic anchors to obtain correlation analysis results; if the correlation analysis results are correlated, identifying emotional changes according to the emotional tendencies and the current user data; adjusting the importance weight of the event units in a life story according to the emotional changes; and updating the presentation content of the event units according to the current user data and the importance weight. The application can extract events with data value by analyzing emotional changes, so that life course data analysis is realized, and the data value extraction capability is improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a life course data analysis method and system. Background Technology

[0002] In the digital age, the amount of personal digital information is exploding, existing in various forms. Integrating and processing this scattered and diverse personal information can help users construct a coherent and meaningful review of their personal life journey. Existing methods access multiple user-specified digital repositories, sort them by timestamps, perform keyword recognition on text content, and conduct basic object recognition and scene analysis on images and videos to initially classify and label the information, forming a raw timeline. However, with the increasing frequency of users' digital activities, this data contains a large amount of mundane, repetitive, or narrative-lacking "noise," causing truly significant or pivotal events to be submerged in the sea of ​​information. The resulting life story is often a lengthy, detailed, and unfocused chronicle, leaving users feeling overwhelmed and unable to extract valuable personal journey details, easily leading to reading fatigue. Existing methods have weak data value extraction capabilities.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this invention is to propose a life course data analysis method and system that can extract data-valuable events by analyzing emotional changes, thereby achieving life course data analysis and improving the ability to extract data value.

[0005] On one hand, embodiments of the present invention provide a life course data analysis method, including the following steps: Obtain historical user data; The historical user data is clustered into event units. Feature extraction is performed on the event units to obtain core elements and emotional tendencies; Based on the core elements and the emotional tendency, semantic anchors for the event units are generated, and the representation of the semantic anchors includes keyword combinations and embedding vectors; Perform a correlation analysis on the current user data and the semantic anchor points to obtain the correlation analysis results; If the correlation analysis result indicates a correlation, then based on the sentiment tendency and the current user data, the sentiment change is identified; Adjust the importance weight of the event unit in the life story based on the emotional changes; The presentation content of the event unit is updated based on the current user data and the importance weight.

[0006] In some embodiments, identifying emotional changes based on the emotional tendency and the current user data includes: The current user data is deconstructed to identify functional segments, which are used to represent factual restatements, causal attributions, and impact assessments of the event unit. Perceptual information is obtained by performing sensory extraction on the aforementioned functional segments; Based on the historical user data, establish a personalized user experience expression pattern; The sensory information is calibrated according to the sensory expression pattern. Based on the functional segments, logical relationships, and calibrated perceptual information, a semantic structure vector is constructed; The emotional changes are identified based on the semantic structure vector and the emotional tendency.

[0007] In some embodiments, the step of deconstructing the current user data and identifying functional segments includes: Identify functional expressions in the current user data, including metaphorical expressions, emotional words, or scene descriptions; If the functional expression is a metaphorical expression, then the functional fragment is identified based on the historical user data; If the functional expression is an emotional word, then the functional segment is identified based on the context of the emotional word, and the context is used to determine whether the emotional word directly expresses feelings. If the functional expression is a scene description, then the event background, cause of occurrence and subsequent impact contained in the scene description are analyzed to identify the functional segment.

[0008] In some embodiments, identifying the functional segment based on the historical user data includes: Identify the vehicle of the metaphorical expression; Determine the core semantic meaning of the metaphor in the current context; Based on the historical user data, identify metaphorical semantic mapping rules; Based on the core semantic orientation and the metaphorical semantic mapping rules, identify the semantic association between the metaphor and the event ontology; The confidence level of the semantic association is evaluated to obtain the confidence evaluation result; Based on the confidence assessment results, the facts referred to by the metaphorical expression are determined, and the functional fragment is obtained.

[0009] In some embodiments, when the current user data contains non-textual data, the step of deconstructing the current user data and identifying functional segments includes: Multimodal information extraction is performed on the non-textual data to obtain visual, auditory, and motion information; The visual information is analyzed to obtain scene content, which includes objects, people, actions, and environment. The auditory information is subjected to sound event recognition to obtain sound events, which include speech, music and ambient sound; The motion information is analyzed to obtain behavioral patterns, which include walking, running, jumping, sitting, and waving. The functional segments are identified based on the scene content, the sound events, the behavioral patterns, and the semantic anchors.

[0010] In some embodiments, the step of performing scene content analysis on the visual information to obtain scene content includes: The visual information is preprocessed to remove noise or enhance image features; Multi-scale analysis is performed on the preprocessed visual information to identify the features of visual elements at different scales; Spatial relationship analysis is performed on the visual element features to determine the relative positions and occlusion relationships between visual elements, and multi-scale analysis results are obtained. Based on the multi-scale analysis results, the occluded or incomplete visual elements are inferred to complete the inference results. Based on the multi-scale analysis results and the completion inference results, the scene content is identified.

[0011] In some embodiments, the step of performing sound event recognition on the auditory information to obtain sound events includes: The auditory information is converted into the time-frequency domain to obtain a time-frequency representation of the auditory information; Noise suppression and reverberation cancellation are performed on the time-frequency representation of the auditory information to obtain a clean time-frequency representation; The pure time-frequency representation is subjected to source separation to obtain the source components; The acoustic features are obtained by extracting features from the sound source components. Based on the acoustic characteristics, the sound source components are classified to obtain the sound events.

[0012] In some embodiments, the step of performing behavioral pattern analysis on the motion information to obtain a behavioral pattern includes: The motion information is smoothed to eliminate random jitter in the motion trajectory; Outlier removal is performed on the smoothed motion information to remove data points that do not conform to normal motion patterns; Noise filtering is applied to the motion information after outlier removal to reduce the impact of environmental interference on posture change recognition. Motion features are extracted from the noise-filtered motion information, including velocity, acceleration, angular velocity, and joint angles. Based on the motion characteristics, the motion information is classified to obtain the behavior pattern.

[0013] In some embodiments, the smoothing process for the motion information includes: Based on the local features of the motion information, determine the smoothing window length threshold and the smoothing intensity threshold; When the motion trajectory changes drastically, a first smoothing window length and a first smoothing intensity are set, wherein the first smoothing window length is less than the smoothing window length threshold and the first smoothing intensity is less than the smoothing intensity threshold. The motion information is smoothed based on the length of the first smoothing window and the first smoothing intensity to preserve motion details; When the motion trajectory changes smoothly, a second smoothing window length and a second smoothing intensity are set, wherein the second smoothing window length is greater than the smoothing window length threshold and the second smoothing intensity is greater than the smoothing intensity threshold. The motion information is smoothed according to the second smoothing window length and the second smoothing intensity to eliminate random jitter.

[0014] On the other hand, embodiments of the present invention provide a life course data analysis system, including: The data acquisition module is used to acquire historical user data; The event clustering module is used to cluster the historical user data into event units. The feature extraction module is used to extract features from the event unit to obtain core elements and emotional tendencies; The semantic anchor generation module is used to generate semantic anchors for the event unit based on the core elements and the sentiment tendency. The semantic anchors are represented by keyword combinations and embedding vectors. The correlation analysis module is used to perform correlation analysis on the current user data and the semantic anchor points to obtain the correlation analysis results; The sentiment analysis module is used to identify sentiment changes based on the sentiment tendency and the current user data if the correlation analysis result indicates a correlation. The weighting adjustment module is used to adjust the importance weight of the event unit in the life story based on the emotional changes. The story update module is used to update the presentation content of the event unit based on the current user data and the importance weight.

[0015] The embodiments of this application include at least the following beneficial effects: First, historical user data is acquired, and event clustering is performed on the historical user data to obtain event units. Then, feature extraction is performed on the event units to obtain core elements and sentiment tendencies, and semantic anchors of the event units are generated. Next, correlation analysis is performed on the current user data and semantic anchors to obtain correlation analysis results. If the correlation analysis results show that there is a correlation, then sentiment changes are identified based on sentiment tendencies and current user data. Finally, the importance weight of the event unit in the life story is adjusted, and the presentation content of the event unit is updated based on the current user data and importance weights. Thus, it is possible to extract events with data value by analyzing sentiment changes, so as to realize life course data analysis and improve the data value extraction capability.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and the drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. 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 a life course data analysis method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a life course data analysis system according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0020] In the digital age, people leave behind a vast amount of personal information in their daily lives. This information exists in various forms, such as text records on social media platforms, photos and videos in personal albums, schedules in calendar applications, and even geolocation information recorded by smart devices. It is necessary to effectively integrate and process this scattered and diverse personal information to build a coherent and meaningful personal life journey review for users.

[0021] Existing methods obtain user authorization to access multiple designated digital repositories, such as social media archives, cloud photo storage, communication logs, and scheduling applications. After data acquisition, initial processing is performed, primarily based on strict chronological sorting according to the timestamps attached to the information, forming a raw timeline. Subsequently, through keyword recognition of text content and basic object recognition and scene analysis of images and videos, this information can be preliminarily classified and labeled, such as identifying rough event categories like "travel," "work meetings," and "family gatherings." The initial intention of this approach is to aggregate scattered digital fragments, providing users with a structured and revisitable overview of their digital lives.

[0022] However, in practice, this approach quickly encountered an unexpected challenge. With the increasing frequency of users' digital activities and the continuous reduction in data storage costs, the amount of personal digital information that the system could access after authorization exploded. This wasn't just a surge in photos and videos, but also included massive amounts of instant messaging records, daily check-ins, web browsing history, and even physiological data recorded by smart wearable devices. While this data was incredibly abundant, the vast majority was mundane, repetitive, or "noise" information lacking clear narrative value. For example, similar landscape photos taken during daily commutes or numerous irrelevant chat logs with friends dominated the data, drowning out truly significant or pivotal events. When attempting to construct life stories, the system found that its output was often a lengthy, detailed, and unfocused chronicle. Users felt overwhelmed by the information and struggled to extract valuable personal journeys, leading to reading fatigue.

[0023] To address this information overload issue, the technical team introduced preliminary data filtering and deduplication mechanisms. The system began using a series of preset rules for data cleaning, such as identifying and removing duplicate photos or video frames, merging large amounts of similar text messages within a short period, or filtering out content based on a preset list of "low-value" keywords. Furthermore, it attempted to initially aggregate or remove overly dense or sparse data points by setting time windows and activity frequency thresholds. For example, if a user checks in multiple times consecutively at the same location, the system merges these into a single "stayed at this location" event and retains only one representative record. This method reduced the amount of data to some extent, making the initially generated timeline less bloated, allowing users to quickly skip over obviously repetitive information while browsing.

[0024] However, this simple rule-based filtering quickly revealed its limitations. Despite the reduced data volume, the system often indiscriminately removed seemingly trivial information that might actually contain important context or emotional clues during the filtering process. For example, a seemingly ordinary daily conversation might contain the seeds of an important decision, but this nascent information was filtered out by the "low-value" rule; a casually taken photo with mediocre composition might be a record of an important person's first appearance in the user's life, but it was ignored by the system because it was "unrefined." The system failed to understand the deeper meanings hidden behind these "low-value" data points or their potential connections within the overall life narrative. This resulted in a life story that, while somewhat concise, appeared flat, lacking depth and emotional coherence. Many key events and details were unintentionally erased, leaving users feeling that the story lacked a "soul" when reviewing it, failing to truly touch their hearts or provide a deeper level of self-awareness.

[0025] To compensate for the information loss caused by simple filtering, engineers began exploring more advanced event recognition and summarization techniques. The system was designed to identify more complex event patterns. For example, by analyzing the correlation between geographical location, participants, time span, and content themes, it clusters a series of scattered data points into an "event unit," such as "a graduation trip in a certain month of a certain year." Based on this, the system attempts to generate a short summary for each identified event and selects the most representative image or text fragment as the event's cover or highlight. This approach aims to extract structured events from massive amounts of data and provide an overview, allowing users to browse by event unit rather than simply viewing scattered data points in chronological order.

[0026] However, new challenges arise when the system attempts to assign "meaning" or "importance" to these events. Even if the system can accurately identify and cluster various events, it still struggles to determine the "personal importance" or "emotional value" of these events to the user. For example, a grand company annual meeting might be a "major event" in terms of data volume and number of participants, and the system would assign it high weight and present it in detail, but for the user, it might just be routine or even boring. On the other hand, a late-night conversation with an old friend might only have a few messages recorded, but it could have a profound impact on the user's life trajectory, yet the system might consider it a minor event due to its small data volume. The system lacks an understanding of the user's subjective feelings and values, and cannot distinguish which events are truly cherished milestones and which are merely fleeting moments in life. This results in a lack of hierarchy in the presentation of the generated life stories; important events are not highlighted enough, while some mundane events may be overemphasized, making the entire narrative unbalanced and failing to truly reflect the user's deepest life experiences.

[0027] Existing methods often focus on identifying objective events and reducing data volume, but they struggle to capture the deep personal meaning of events and cannot adapt to the "meaning reconstruction" of past events by users over time. This results in life stories that are lengthy and lack focus, fail to resonate with users emotionally, and cannot adjust their narrative focus in real time as users' cognition evolves, leading to weak data value extraction capabilities.

[0028] The embodiments of this application will be explained in detail below with reference to the accompanying drawings: Figure 1 This is an optional flowchart of a life course data analysis method provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S108.

[0029] Step S101: Obtain historical user data; Step S102: Perform event clustering on historical user data to obtain event units; Step S103: Extract features from event units to obtain core elements and sentiment tendencies; Step S104: Generate semantic anchors for event units based on core elements and sentiment tendencies. The semantic anchors can be represented by keyword combinations and embedding vectors. Step S105: Perform correlation analysis on the current user data and semantic anchors to obtain the correlation analysis results; Step S106: If the correlation analysis result shows that there is a correlation, then identify the sentiment change based on the sentiment tendency and the current user data; Step S107: Adjust the importance weight of event units in the life story according to emotional changes; Step S108: Update the presentation content of the event unit based on the current user data and importance weight.

[0030] Steps S101 to S108 shown in the embodiments of this application can extract events with data value by analyzing emotional changes, thereby realizing life course data analysis and improving the ability to extract data value.

[0031] In some embodiments, steps S101-S108 may first acquire historical user data. For example, this can be achieved by granting access to public or private data on the user's social media platform; or by integrating local data from the user's personal device (such as a smartphone or smart wearable device), including photo galleries, video recordings, diary applications, health monitoring data, etc.; or by allowing the user to actively upload data files they wish to include in their life story. It is understood that historical user data refers to various types of digital information generated and stored by the user over a past period, such as social media posts, chat logs, photos, videos, diaries, emails, schedules, etc. This data can be structured or unstructured, and together they constitute the digital footprint of the user's personal life trajectory.

[0032] Then, historical user data is clustered to obtain event units. For example, time-series-based clustering methods can be used to group user data that are temporally close and content-related into a single event unit. Specifically, a time window can be set, such as 24 hours or a week, and all relevant data occurring within that time window (such as multiple photos from the same location or chat logs on the same topic) can be clustered into a potential event. Alternatively, content-similarity-based clustering methods can be used. Natural language processing techniques can be used to analyze the semantic similarity of text data, or image recognition techniques can be used to analyze the visual similarity of image data, clustering data that are highly similar semantically or visually. For example, multiple photos containing "birthday cake" and "party hat" can be identified and combined with the related chat logs mentioning "Happy Birthday," thus clustering them into a "Birthday Party" event unit. It can be understood that an event unit refers to an independent event with a specific theme, time frame, and participants, identified through clustering analysis of historical user data. For example, a trip, a birthday party, or an important work project can all be considered an event unit.

[0033] Feature extraction is performed on event units to obtain core elements and sentiment tendencies. For core element extraction, key entities can be identified from the event unit, such as names of people, places, organizations, and time points. For example, information such as "Zhang XX," "XX Ancient City," and "XX Festival" can be extracted from a travel event unit. Simultaneously, key actions or activities within the event unit can be identified, such as "hiking," "taking photos," and "tasting food." For sentiment tendency extraction, sentiment analysis models can be used to determine the sentiment polarity of the text content within the event unit, identifying positive, negative, or neutral emotions. For example, by analyzing words like "wonderful," "pleasant," and "relaxing" used by users in their travel diaries, the sentiment tendency of the travel event can be determined to be positive. Furthermore, facial expression recognition and voice tone analysis from images and videos can be combined to comprehensively determine the sentiment tendency of the event unit. It can be understood that core elements refer to the key information constituting the event unit, such as the time, place, participants, and main activities of the event. Sentiment tendency refers to the overall emotional tone contained within the event unit, such as positive, negative, neutral, joyful, sad, or surprised.

[0034] Based on core elements and sentiment tendencies, semantic anchors are generated for event units. These anchors can be represented through keyword combinations or embedded vectors. For example, for an event unit "Travel to River A," core elements such as "River A" and "food" are extracted, and the sentiment tendency is "pleasure." A semantic anchor can be generated by concatenating these core elements and sentiment tendency, such as "River A - Food - Pleasure," which summarizes the core content and emotional tone of the event. A semantic anchor is a structured, computable representation designed to transform the semantic and sentiment attributes of an event unit into actionable data. A semantic anchor can be a keyword combination or an embedded vector, its purpose being to efficiently represent the semantic and sentiment characteristics of the event unit in subsequent correlation analysis. In essence, a semantic anchor is an abstract representation of an event unit; by integrating core elements and sentiment tendencies, it forms a semantic identifier that represents the deeper meaning and emotional value of the event unit, facilitating subsequent correlation analysis. In one implementation, a semantic anchor can be represented as a keyword combination that directly summarizes the core content and sentiment of the event. For example, "A River - Food - Pleasure" is a keyword combination. This form quantifies semantics primarily through the selection and combination of words. Another implementation method represents semantic anchors as embedding vectors. These vectors, through deep learning models (such as word embeddings, sentence embeddings, or event embeddings), map the core elements and sentiment of event units into a continuous vector space. In this vector space, event units that are semantically and emotionally similar will have their corresponding embedding vectors closer together, or their cosine similarity will be higher. Each dimension of this vector represents a certain abstract feature of the event, thus achieving a numerical representation of semantics and sentiment.

[0035] The system performs correlation analysis on current user data and semantic anchors to obtain the results. For example, when a user posts a new social media update containing the message "I want to go to Ajiang again; the food there is so delicious," the system compares this current user data with the previously generated semantic anchor "Ajiang-food-pleasure." Correlation analysis can be achieved by calculating the semantic similarity between the current user data and the semantic anchor, for example, using a word vector model to calculate text similarity, or using a topic model to determine if they belong to the same topic category. If the similarity exceeds a preset similarity threshold, a correlation is considered to exist. The preset similarity threshold can be calibrated.

[0036] If the correlation analysis indicates a correlation, then the system identifies sentiment shifts based on sentiment tendencies and current user data. For example, if a user expresses "nostalgia" and "longing" for "A River" in the current data, while the sentiment tendency for the previous "A River Trip" event unit was "pleasure," the system may identify a sentiment shift of "continuation and deepening of positive emotions." If a user expresses "regret" or "dissatisfaction" with an event in the current data, while the historical sentiment tendency for that event was "neutral" or "positive," the system may identify a shift of "emotion from neutral / positive to negative."

[0037] Adjust the importance weight of event units within the life story based on changes in sentiment. For example, if a user's sentiment towards an event shifts from "neutral" to "strong nostalgia," it indicates that the event has higher emotional value for the user, and therefore the importance weight of that event unit can be increased. Conversely, if the sentiment change shows a decrease in user attention to an event or a diminishing of sentiment, its weight can be appropriately decreased. Importance weight can be a numerical value used to determine the prominence, display duration, or level of detail of events in the presentation of the life story.

[0038] Finally, the presentation content of the event unit is updated based on the current user data and importance weights. For example, if the importance weight of the "A River Trip" event unit increases due to the user's continued nostalgia, the system may allocate more display space for the event when generating a life story, such as displaying more related photos and videos, or providing a more detailed text description. Simultaneously, the "nostalgia" sentiment expressed in the current user data can also be integrated into the event's presentation content. For example, adding statements like "The user still fondly remembers the food and scenery of A River" to the event description makes the life story presentation more personalized and emotionally profound. This embodiment effectively solves the problems of information overload, lack of depth, and personalization faced by existing methods when processing massive amounts of personal digital information by introducing semantic anchors, emotion change recognition, and importance weight adjustment mechanisms.

[0039] Through the above technical solution, this embodiment, by recognizing semantic anchors and emotional changes, can more accurately determine the value of information and the user's emotional investment, avoiding indiscriminate deletion and thus preserving the integrity and coherence of the story. By dynamically adjusting the importance weights, the life story can adaptively highlight key events according to the user's emotional changes, thereby providing a truly personalized and emotionally resonant review of the life journey. Therefore, this embodiment can provide users with a more in-depth life story that is closer to their inner experience, significantly enhancing the user's sense of value and satisfaction in reviewing their personal journey.

[0040] In some embodiments, step S106, identifying emotional changes based on sentiment tendency and current user data, may include, but is not limited to, the following steps: Step S201: Deconstruct the current user data to identify functional segments. These functional segments are used to represent factual restatements, causal attributions, and impact assessments of event units. Step S202: Extract sensory information from functional segments to obtain sensory information; Step S203: Based on historical user data, establish a personalized user experience expression pattern; Step S204: Calibrate the sensory information according to the sensory expression pattern; Step S205: Construct a semantic structure vector based on functional fragments, logical relationships, and calibrated perceptual information; Step S206: Identify emotional changes based on semantic structure vectors and sentiment tendencies.

[0041] In some embodiments, the current user data can be deconstructed first to identify functional segments. A functional segment can be a structured representation used to represent a user's factual restatement, causal attribution, and impact assessment of an event unit. These functional segments can include factual restatements of the event unit, such as a user describing the specific details of an event: "This photo was taken last summer in Ajiang Ancient Town. The sun was shining brightly, and we walked along the stone path for a long time." The information such as "this photo was taken last summer in Ajiang Ancient Town," "the sun was shining brightly," and "walked along the stone path for a long time" constitutes a factual restatement of the event unit "Ajiang Trip." The attribution of the event's causes can also be included, such as a user explaining the internal or external factors that led to the event: "That startup failed mainly because of misjudgment of the market and lack of experience." The information such as "misjudgment of the market" and "lack of experience" constitutes a causal attribution of the "startup failure" event. And the assessment of the subsequent impact of the event, such as users expressing the long-term impact of the event on their life or emotions, such as "Although that experience was difficult, it also helped me grow a lot, and now I understand the importance of perseverance." Information such as "greatly grew" and "understands the importance of perseverance" constitutes an impact assessment of the "difficult experience" event. Each functional segment can be encoded as structured data containing information such as its type, content summary, and related entities (e.g., factual restatements, causal attributions, and impact assessments). Through content deconstruction, user feedback on the event can be understood from multiple dimensions.

[0042] Then, the functional segments are processed to extract the emotional information. This information can be used through natural language processing techniques or sentiment analysis models to identify and quantify the emotions, feelings, or attitudes expressed by users in these segments. This emotional information can be positive, negative, or neutral, and can be further subdivided into specific emotional categories such as joy, sadness, anger, and surprise.

[0043] Then, based on historical user data, a personalized emotional expression pattern is established for each user. This can be achieved by analyzing a user's past text, voice, or behavioral data to identify their emotional expression habits, vocabulary preferences, and tone characteristics, thus constructing a pattern that reflects their unique emotional expression style. For example, some users may be accustomed to expressing emotions subtly, while others may be more direct. It is understood that a emotional expression pattern refers to a pattern established based on a user's past text, voice, or behavioral data that reflects that user's unique emotional expression habits and methods. In lifecycle data analysis methods, because different users may express the same emotions in significantly different ways (for example, some users are accustomed to direct expression, while others tend to use metaphors, irony, or subtle wording), establishing a personalized emotional expression pattern aims to more accurately understand the emotions expressed by users in specific contexts, thereby improving the accuracy of subsequent emotional information calibration. One implementation method is to establish an emotional expression pattern by analyzing the vocabulary, phrases, and sentence structures that users tend to use when expressing specific emotions. For example, some users might frequently use "not bad" to express positive emotions, while others might use "fantastic" to express positive emotions. The mapping relationship between these personalized words and emotions can be recorded as a pattern of emotional expression. Another approach is to identify users' frequently used rhetorical devices such as metaphors, hyperbole, and irony, and their application patterns in different emotional expressions, as a pattern of emotional expression. For example, if a user repeatedly uses irony to express dissatisfaction in historical data, their emotional expression pattern can be represented as using irony to express dissatisfaction. In subsequent applications, if ironic expressions appear in the current data, they can be interpreted as negative emotions.

[0044] Based on the pattern of emotional expression, emotional information is calibrated to eliminate misjudgments caused by individual differences in expression, making the identification of users' true emotions more accurate. For example, a user who habitually uses irony may have their seemingly negative expression identified as a positive emotion after personalized calibration.

[0045] A semantic structure vector is constructed based on functional fragments, logical relationships, and calibrated perceptual information. Discrete functional fragments, logical relationships, and calibrated perceptual information can be integrated to form a coherent and complete semantic structure vector. Logical relationships can include causal, adversative, and parallel relationships, which help reveal the deep connections between different fragments, thus more comprehensively reflecting the user's overall cognition and emotional experience of the event. Understandably, logical relationships can include causal relationships (e.g., a cause leads to a fact or effect), adversative relationships (e.g., an emotional shift between a fact and an effect), and parallel relationships. These can be represented using graph structures, semantic networks, or logical expressions. This structured representation enables the system to understand the dependencies and flow between fragments, thereby analyzing meaning at a higher level. By analyzing these logical relationships, a deeper understanding of the coherence and deeper meaning of the user's narrative can be achieved. Calibrated perceptual information is information extracted from the user's emotions or attitudes from the functional fragments and calibrated according to the user's personalized perceptual expression patterns. This can be quantified as an emotional polarity score (e.g., a continuous value from -1 to 1, representing a range from negative to positive), an emotional intensity level (e.g., an intensity level from 1 to 5), or a vector representation in a multidimensional emotional space (e.g., pleasure, activation, dominance, etc.). This quantification allows for the comparison and computation of emotional information. The semantic structure vector obtained by integrating functional fragments, logical relationships, and calibrated emotional information can comprehensively reflect the user's overall cognition and emotional experience of an event, revealing deep connections between different fragments, thus providing a more insightful view of the user's current state.

[0046] Finally, based on the semantic structure vector and sentiment tendency, the change in sentiment is identified. For example, if the initial sentiment tendency of an event unit is positive, but the semantic structure vector reflects the user's negative attribution to the event and feelings of sadness, then the change in sentiment from positive to negative, or a decrease in the degree of positivity, can be identified.

[0047] Through the above technical solution, this embodiment introduces the identification of functional segments and the calibration of personalized user expression patterns, enabling the identification of emotional changes to move beyond simple keyword matching or preset rules. Instead, it delves deeper into the underlying meanings and individual differences expressed by users. This significantly improves the accuracy and robustness of emotional change identification, ensuring that adjustments to the importance weights of events in a user's life story and updates to the presented content are more closely aligned with the user's true feelings and perceptions, thereby enhancing the personalization and emotional resonance of the life story.

[0048] In some embodiments, step S201, which involves deconstructing the current user data and identifying functional segments, may include, but is not limited to, the following steps: Step S301: Identify functional expressions in the current user data. Functional expressions include metaphorical expressions, emotional words, or scene descriptions. Step S302: If the functional expression is a metaphorical expression, then identify the functional fragment based on historical user data; Step S303: If the functional expression is an emotional word, then the functional segment is identified based on the context of the emotional word. The context is used to determine whether the emotional word directly expresses feelings. Step S304: If the functional expression is a scene description, then analyze the event background, cause of occurrence and subsequent impact contained in the scene description to identify functional segments.

[0049] In some embodiments, functional expressions in the current user data can be identified first. Functional expressions refer to linguistic expressions in the current user data that instruct or imply a user's factual restatement, causal attribution, or impact assessment of a specific event unit. These expressions can be any combination of one or more of metaphorical expressions, emotional vocabulary, or scene descriptions. For example, metaphorical expressions may indirectly describe the event through imagery, emotional vocabulary directly expresses the user's emotional state, and scene descriptions convey the details and impact of the event through the depiction of a specific context.

[0050] If the functional expression is metaphorical, then the functional segment is identified based on historical user data. Metaphorical expressions are often personalized and context-dependent. By analyzing users' past metaphorical usage habits and the facts they refer to, the deeper meaning of the current metaphorical expression can be understood more accurately, thereby identifying the corresponding functional segment.

[0051] If the functional expression is an emotional word, then the functional segment is identified based on the context of the emotional word. The context is used to determine whether the emotional word directly expresses feelings. Contextual analysis is crucial because the same emotional word may express different feelings in different contexts, or may not even directly express feelings (e.g., irony). Through a deep understanding of the context, it is possible to determine whether the emotional word truly and directly reflects the user's feelings about the event unit, thereby identifying the corresponding functional segment.

[0052] If the functional expression is a scene description, then the event background, cause, and subsequent impact contained in the scene description are analyzed to identify functional segments. Scene descriptions typically contain rich narrative information. By deconstructing this information, the specific context in which the event occurred, the causes that led to the event, and the impact of the event on the user or the surrounding environment can be extracted. These are all key elements constituting a functional segment.

[0053] Through the above technical solution, this embodiment can significantly improve the accuracy and comprehensiveness of functional segment recognition. By introducing the concept of functional expressions and providing customized recognition strategies for metaphorical expressions, emotional vocabulary, and scene descriptions, these challenges are effectively overcome. This enables the system to gain a deeper understanding of the nuances of user expression, thereby providing more reliable and richer information for subsequent emotional change recognition, and ultimately improving the overall accuracy of life course data analysis and user experience.

[0054] In some embodiments, in step S302, identifying functional segments based on historical user data may include, but is not limited to, the following steps: Identify the vehicle in a metaphorical expression; Determine the core semantic meaning of the metaphor in the current context; Identify metaphorical semantic mapping rules based on historical user data; Based on the core semantic orientation and metaphorical semantic mapping rules, identify the semantic relationship between the metaphor and the event ontology; The confidence level of semantic associations is evaluated to obtain the confidence evaluation results; Based on the confidence assessment results, the facts referred to by the metaphorical expression are determined, and the functional fragment is obtained.

[0055] In some embodiments, the vehicle of a metaphorical expression can be identified first. For example, the specific words or phrases used to refer to or symbolize other things or concepts can be accurately identified from the metaphorical expression. For instance, in "Life is a journey," "journey" is the vehicle. Further, the core semantic meaning of the vehicle in the current context is determined, aiming to analyze the deeper meaning or key attributes that the vehicle emphasizes in a specific context. For example, in the example of "Life is a journey," the core semantic meaning of "journey" might be "process," "experience," or "exploration."

[0056] Then, based on historical user data, metaphorical semantic mapping rules are identified. By leveraging users' past language habits and expression patterns, a correspondence between the metaphor and its referent event can be constructed, serving as the metaphorical semantic mapping rules. These rules can be personalized, reflecting users' unique ways of using metaphors. For example, by analyzing historical data on users' use of the metaphor "travel" in different contexts, it can be discovered that it may have a mapping relationship with event events such as "life stage transitions" or "challenges and growth." Furthermore, based on the core semantic orientation and metaphorical semantic mapping rules, the semantic association between the metaphor and the event event is identified, thereby linking abstract metaphorical expressions with specific events or facts.

[0057] Next, a confidence assessment is performed on the semantic association to obtain the confidence assessment result. This assessment process can comprehensively consider the strength of the metaphorical semantic mapping rule and the clarity of the current context to quantify the reliability of the semantic association. For example, if a metaphorical semantic mapping rule frequently appears in historical data and highly matches the current context, its confidence assessment result will be high. Understandably, the strength of a metaphorical semantic mapping rule refers to the frequency and consistency of its appearance in the user's historical data. If the mapping relationship between a metaphor and a specific event ontology appears frequently and stably in the user's historical expressions, then the strength of the metaphorical semantic mapping rule is higher, and the confidence of the semantic association identified in the current context will also increase accordingly. For example, if a user has repeatedly used "climbing a peak" as a metaphor for "overcoming career challenges" in the past, then when they use "climbing a peak" again, the system will consider its semantic association with "career challenges" to be of high strength. The clarity of the current context refers to the degree to which the context of the metaphorical expression in the current user data explicitly supports the metaphorical semantic mapping rule. If other words, phrases, and themes in the current context are highly consistent with the event ontology pointed to by the metaphorical semantic mapping rule, then the confidence of the semantic association will be higher. For example, if a user mentions words such as "project," "team," and "difficulty" when describing "climbing a mountain," this contextual information will enhance the confidence of the semantic association between "climbing a mountain" and "career challenge."

[0058] Finally, based on the confidence assessment results, the facts referred to by the metaphorical expression are determined, resulting in a functional fragment. This means that when the confidence of semantic association reaches a preset confidence threshold, the factual information implied by the metaphorical expression can be extracted as a functional fragment for subsequent processing. The preset confidence threshold can be calibrated.

[0059] Through the above technical solution, this embodiment can effectively improve the depth and accuracy of understanding metaphorical expressions in user data. By utilizing users' personalized historical data, it can more accurately capture users' unique expression habits and metaphorical meanings, thereby significantly improving the reliability of extracting factual functional fragments from complex language. This precise grasp of the nuances of language helps to construct more authentic and detailed life stories, and provides a solid foundation for subsequent sentiment recognition and importance weight adjustment, thereby improving the robustness and intelligence level of the entire life course data analysis method.

[0060] In some embodiments, in step S201, when the current user data contains non-textual data, the current user data is deconstructed to identify functional segments, which may include, but is not limited to, the following steps: Step S401: Extract multimodal information from non-textual data to obtain visual, auditory, and motion information; Step S402: Perform scene content analysis on the visual information to obtain scene content, which includes objects, people, actions, and environment; Step S403: Perform sound event recognition on the auditory information to obtain sound events, which include speech, music and ambient sound; Step S404: Perform behavioral pattern analysis on the motion information to obtain behavioral patterns, which include walking, running, jumping, sitting, and waving. Step S405: Identify functional segments based on scene content, sound events, behavioral patterns, and semantic anchors.

[0061] In some embodiments, user data may contain rich non-textual data, such as images, videos, and audio, which also carry important emotional and event information. Relying solely on text analysis may fail to comprehensively and accurately capture the user's authentic experiences and emotional changes within their life story, leading to incomplete or inaccurate functional segment identification. Therefore, multimodal information extraction can be performed on the non-textual data to obtain visual, auditory, and motion information. For example, inherent, analyzable feature information can be extracted from these different forms of data using appropriate processing techniques. For instance, for video data, visual information (image content), auditory information (sound), and motion information (movement of people or objects) can be extracted. It is understood that non-textual data refers to data that does not directly express information in the form of text or symbols, such as images, videos, audio, and biosensor data.

[0062] Then, scene content analysis is performed on the visual information to obtain scene content. This aims to identify and understand specific elements and their interrelationships within the visual information using image processing and computer vision techniques. For example, it can identify specific objects (such as gifts or trophies), people (such as family or friends), their ongoing actions (such as celebrating or hugging), and the environment (such as a birthday party or graduation ceremony). Scene content includes objects, people, actions, and environment; visual information refers to the visual features of the scene, objects, people, actions, and environment obtained from images or videos.

[0063] Subsequently, sound event recognition is performed on the auditory information to obtain sound events. This aims to classify and understand the auditory information, identifying the speech (such as dialogue content), music (such as the type and mood of background music), and ambient sounds (such as applause, crying, and laughter). These sound events can provide important clues for understanding the user's emotional state and the context of the event. Sound events include speech, music, and ambient sounds; auditory information refers to the sound features extracted from the audio data.

[0064] Next, behavioral pattern analysis is performed on the motion information to obtain behavioral patterns. The aim is to identify specific behavioral patterns, such as walking, running, jumping, sitting, and waving, by processing motion information. These behavioral patterns are often associated with specific events or emotional states; for example, jumping may indicate excitement, and sitting may indicate calmness or contemplation. Specifically, behavioral patterns include walking, running, jumping, sitting, and waving, while motion information refers to the characteristics of the trajectory and posture changes of people or objects obtained from video or sensor data.

[0065] Finally, based on scene content, sound events, behavioral patterns, and semantic anchors, functional segments representing factual restatements, causal attributions, and impact assessments in the current user data can be identified more comprehensively and accurately. Semantic anchors play a guiding and associative role in this process, helping the system match and understand multimodal information with known event units.

[0066] To illustrate this technical solution more clearly, a specific example is used below. Assume the current user data is a user-uploaded video recording their graduation ceremony. The system first performs multimodal information extraction on the video. The visual information extraction module identifies scene content such as graduation caps, gowns, crowds, and the stage, as well as actions like the user hugging and waving with family. The auditory information extraction module identifies the background graduation march, cheers from the crowd, and snippets of dialogue between the user and their family. The motion information extraction module analyzes behavioral patterns such as the user's walking posture and body language during hugs. Subsequently, these extracted scene content, sound events, and behavioral patterns are correlated with the semantic anchor point of the event unit "graduation ceremony joy." For example, based on the identified graduation caps, gowns, cheers, and hugs, the system determines that these non-textual elements collectively point to the event "graduation ceremony," and that the cheers and hugs strongly express the emotion of joy. Therefore, even without explicit text descriptions in the video, the system can accurately identify functional segments such as "factual retelling of the graduation ceremony," "joy arising from graduation," and "impact assessment of anticipation for the future," thereby recognizing the user's emotional changes and adjusting the weight and presentation of that event unit in the life story accordingly.

[0067] Through the above technical solution, this embodiment can effectively process current user data containing non-textual data, greatly expanding the applicability of life course data analysis methods. This embodiment can comprehensively capture users' emotional expressions and event details from multiple dimensions such as vision, hearing, and motion, significantly improving the accuracy and completeness of functional segment recognition. This allows the system to deeply understand user experience even with diverse user expression methods, thereby more accurately identifying emotional changes and adjusting the importance weight of event units accordingly, ultimately presenting a life story that is closer to the user's real feelings and experiences.

[0068] In some embodiments, step S402 involves performing scene content analysis on the visual information to obtain scene content, which may include, but is not limited to, the following steps: Visual information is preprocessed to remove noise or enhance image features; Multi-scale analysis is performed on the preprocessed visual information to identify the features of visual elements at different scales; Spatial relationship analysis of visual element features is performed to determine the relative positions and occlusion relationships between visual elements, and multi-scale analysis results are obtained. Based on the results of multi-scale analysis, the occluded or incomplete visual elements are inferred to complete the inference results. Based on the results of multi-scale analysis and completion inference, the scene content is identified.

[0069] In some embodiments, visual information can be preprocessed to eliminate random noise in the image by applying a series of image processing techniques, such as noise filtering (e.g., Gaussian filtering, median filtering), or to enhance image features through contrast enhancement, edge detection, etc. The aim is to improve the quality of the visual information, making it more suitable for subsequent analysis and processing. Multi-scale analysis of the preprocessed visual information can then be performed to identify visual element features at different scales. Techniques such as image pyramids, scale-space representation, or multi-resolution analysis can be used to detect and analyze visual information at different spatial resolutions. In this way, visual element features at different granularities can be identified, such as subtle textures, edges, and larger object outlines, thereby comprehensively capturing the structural information in the visual information.

[0070] Then, spatial relationship analysis is performed on the visual element features to determine the relative positions (e.g., above, below, left, right) and occlusion relationships between visual elements. This can be achieved by analyzing the overlapping bounding boxes, depth information, or semantic segmentation results of visual elements. Thus, a multi-scale analysis result containing visual elements and their spatial layout can be obtained.

[0071] Based on the multi-scale analysis results, occluded or incomplete visual elements are then inferred to complete them, yielding the completion inference results. When objects or figures in a scene are partially occluded or only partially displayed, the contextual information provided by the multi-scale analysis results, along with pre-stored knowledge models (e.g., shape models of common objects or figures), can be used to infer the complete form or existence of these incomplete elements. This allows for the acquisition of completion inference results to compensate for the lack of visual information.

[0072] Finally, based on the results of multi-scale analysis and inference, the scene content is identified. By integrating all visible and inferred visual elements, the features of visual elements, and the spatial relationships between visual elements, the specific content in the scene can be accurately identified, including existing objects, people, ongoing actions, and the environment.

[0073] Through the above technical solutions, this embodiment can effectively improve the accuracy and robustness of scene content analysis of visual information in non-textual data. By employing multi-scale analysis and completion inference mechanisms, it overcomes the limitations of traditional methods in handling complex scenes, lighting changes, or partial occlusion. This ensures that even with incomplete or interfering visual information, key elements such as objects, people, actions, and the environment in the scene can be accurately identified. This provides a more reliable and comprehensive visual basis for subsequent functional segment recognition, thereby improving the overall accuracy of life course data analysis and user experience.

[0074] In some embodiments, step S403, which involves recognizing sound events from auditory information to obtain sound events, may include, but is not limited to, the following steps: The auditory information is transformed into a time-frequency domain to obtain a time-frequency representation of the auditory information; Noise suppression and reverberation elimination are performed on the time-frequency representation of auditory information to obtain a clean time-frequency representation; Source separation is performed on the pure time-frequency representation to obtain the source components; Feature extraction is performed on the sound source components to obtain acoustic features; Based on acoustic characteristics, sound source components are classified to obtain sound events.

[0075] In some embodiments, the auditory information can first be transformed into the time-frequency domain to obtain a time-frequency representation of the auditory information. For example, the original auditory information can be transformed from the time domain to the time-frequency domain to facilitate simultaneous analysis of its characteristics in time and frequency. For instance, methods such as Short-Time Fourier Transform (STFT), Wavelet Transform, or Constant Q Transform can be used to decompose the auditory signal into a series of components at different times and frequencies, thereby obtaining a time-frequency representation of the auditory information. The purpose is to reveal the time-varying frequency components in the auditory information, providing a foundation for subsequent noise processing and feature extraction.

[0076] Then, noise suppression and reverberation cancellation are performed on the time-frequency representation of auditory information to obtain a clean time-frequency representation. This aims to improve the quality of the time-frequency representation of auditory information and remove the interference of environmental noise and spatial reverberation on sound event recognition. Specifically, noise suppression can be achieved through spectral subtraction, Wiener filtering, or deep learning noise reduction models to reduce the influence of background noise; reverberation cancellation can be achieved through cepstral mean compensation, blind source separation, or dereverberation algorithms based on room acoustic models to eliminate the tailing effect caused by sound reflection in space. Through these processes, a cleaner time-frequency representation can be obtained, thereby improving the accuracy of subsequent analysis.

[0077] Next, source separation is performed on the pure time-frequency representation to obtain source components. This separates multiple mixed sound sources in the pure time-frequency representation into independent source components. For example, when an auditory message contains human voices, music, and ambient sounds simultaneously, source separation technology can distinguish these different sound components. This can be achieved through methods such as Independent Component Analysis (ICA), Non-negative Matrix Factorization (NMF), or deep learning models (such as the U-Net architecture). The aim is to decompose complex auditory scenes into independently analyzable units, laying the foundation for accurate identification of each sound event.

[0078] Finally, feature extraction is performed on the sound source components to obtain acoustic features. Representative acoustic features can be extracted from the separated sound source components. These acoustic features can effectively describe the essential properties of sound, such as Mel-frequency cepstral coefficients (MFCC), fundamental frequency (pitch), energy, zero-crossing rate, spectral centroid, and spectral bandwidth. These features can quantify the perceptual attributes of sound, such as timbre, pitch, and loudness. The purpose is to transform the high-dimensional raw sound source component data into low-dimensional and discriminative feature vectors, facilitating subsequent classification processing. Based on the acoustic features, the sound source components are classified to obtain sound events. The classification process can utilize machine learning or deep learning models, such as Support Vector Machines (SVM), Gaussian Mixture Models (GMM), Convolutional Neural Networks (CNN), or Recurrent Neural Networks (RNN). These models establish a mapping relationship between features and sound event categories by learning a large amount of labeled acoustic feature data. For example, sound source components can be classified into predefined sound event types such as speech, music, and ambient sounds (e.g., birdsong, car sounds, water sounds), thereby achieving semantic understanding of auditory information.

[0079] Through the above technical solution, this embodiment enables refined identification of sound events in auditory information. A series of steps, including time-frequency domain transformation, noise suppression, sound source separation, and feature classification, significantly improve the accuracy and robustness of sound event recognition. Specifically, noise suppression and reverberation cancellation ensure high-quality, pure sound signals even in complex acoustic environments, avoiding misidentification; sound source separation effectively solves the problem of multi-source aliasing, allowing each independent sound event to be accurately captured. This enables a more precise understanding of auditory content in non-textual data, such as distinguishing background music, human dialogue, or specific environmental sounds, thus providing richer and more accurate semantic cues for the identification of functional segments. This refined sound event recognition capability further enhances the depth and breadth of the deconstruction of current user data content, making the identification of emotional changes and the adjustment of importance weights in life stories more accurate, ultimately presenting life story content that is closer to the user's real experience.

[0080] In some embodiments, step S404 involves performing behavioral pattern analysis on the motion information to obtain a behavioral pattern, which may include, but is not limited to, the following steps: Step S501: Smooth the motion information to eliminate random jitter in the motion trajectory; Step S502: Remove outliers from the smoothed motion information to remove data points that do not conform to normal motion patterns; Step S503: Perform noise filtering on the motion information after outlier removal to reduce the impact of environmental interference on posture change recognition. Step S504: Extract motion features from the noise-filtered motion information. The motion features include velocity, acceleration, angular velocity, and joint angle. Step S505: Classify the motion information according to the motion characteristics to obtain the behavior pattern.

[0081] In some embodiments, the raw motion information often contains random jitter, outliers, and environmental noise, which can reduce the accuracy of behavior pattern analysis and thus affect the recognition accuracy of functional segments. It may also lead to misjudgments of user behavior, thereby affecting the adjustment of importance weights for event units in the life story and the updating of presented content. Therefore, motion information can be smoothed first to eliminate random jitter in the motion trajectory, making the trajectory more continuous and natural. For example, moving average filtering, Gaussian filtering, or Kalman filtering methods can be used to process the motion information.

[0082] Secondly, outlier removal is performed on the smoothed motion information to remove data points that do not conform to normal motion patterns. These outliers may be caused by sensor malfunctions, temporary occlusion, or data transmission errors, and if not removed, they may seriously interfere with subsequent behavior pattern analysis. Statistical methods (such as Z-score, IQR) or distance-based methods (such as local anomaly factors) can be used to detect and remove outliers.

[0083] Secondly, noise filtering is applied to the motion information after outlier removal to reduce the impact of environmental interference on posture change recognition. Environmental interference may include background motion, lighting changes, or the motion of non-target objects, which may introduce low-frequency or mid-frequency noise into the motion information. Low-pass filters or wavelet transforms can be used to filter out this noise, thereby obtaining cleaner motion data.

[0084] Subsequently, motion features are extracted from the noise-filtered motion information. These features include velocity, acceleration, angular velocity, and joint angles. These motion features are key indicators describing motion states and patterns. Velocity represents the rate of change in position, acceleration represents the rate of change in velocity, angular velocity represents the rate of change in rotation, and joint angles directly reflect changes in human posture. Extracting these features helps transform the raw motion data into a more semantic representation, facilitating subsequent classification processing.

[0085] Finally, based on motion characteristics, the motion information is classified to obtain behavioral patterns. The aim is to map the extracted motion features to predefined behavior categories, such as walking, running, jumping, sitting, and waving. Machine learning algorithms such as support vector machines, decision trees, neural networks, or deep learning models can be used to train and classify the motion features, thereby accurately identifying the user's behavioral patterns.

[0086] To illustrate this technical solution more clearly, a specific example is used below. Suppose a user wears a smart bracelet that continuously collects the user's three-axis acceleration and angular velocity data as motion information. During daily activities, the bracelet records a large amount of raw motion data. First, this raw motion information is smoothed. For example, a 5-point moving average filter can be used to average the continuous acceleration and angular velocity data to eliminate random fluctuations caused by slight hand tremors or instantaneous sensor errors. After smoothing, the motion trajectory appears smoother. Second, outlier removal is performed on the smoothed motion information. For example, a normal range can be set; if the acceleration or angular velocity value at a certain point in time exceeds three standard deviations of the normal range within a short period, it is marked as an outlier and replaced (e.g., by filling with the average of adjacent normal data points). This effectively removes extreme data points caused by accidental impacts to the bracelet or data transmission errors.

[0087] Next, noise filtering is applied to the motion information after outlier removal. For example, a low-pass filter with a cutoff frequency of 5Hz can be applied to filter out low-frequency interference that may be caused by environmental vibrations or background noise, ensuring that the motion data mainly reflects the user's own posture changes. Subsequently, motion features are extracted from the smoothed, outlier-removed, and noise-filtered motion information. For example, the average speed per second, maximum acceleration, average angular velocity of the joint (wrist), and motion energy within a specific time window can be calculated. These features can quantify the user's motion state. Finally, based on these extracted motion features, the motion information is classified using a pre-trained classification model (e.g., a deep learning-based recurrent neural network model). For example, when continuous, high-frequency arm and leg movements are detected, the system classifies it as a "running" behavior pattern; when prolonged low-speed or stationary states are detected, accompanied by specific posture features, the system classifies it as a "sitting" behavior pattern. Through these preprocessing steps, the system can more accurately distinguish whether a user is "walking" or "running," or "sitting" or "waving," thus providing high-quality input for subsequent identification of user behavior and emotional changes in specific events.

[0088] Through the above technical solutions, this embodiment effectively purifies the original data and reduces the negative impact of data quality issues on subsequent analysis by smoothing motion information, removing outliers, and filtering noise. As a result, the extracted motion features are more representative, making the classification and recognition of behavioral patterns more accurate, thus enabling more precise identification of functional segments. This improvement not only enhances the performance of life course data analysis methods when processing multimodal data but also provides reliable input for more accurately identifying user emotional changes, adjusting the importance weights of event units, and updating presented content, ultimately making the construction and presentation of life stories more authentic and personalized.

[0089] In some embodiments, the smoothing process for motion information in step S501 may include, but is not limited to, the following steps: Based on the local features of the motion information, determine the smoothing window length threshold and the smoothing intensity threshold; When the motion trajectory changes drastically, a first smoothing window length and a first smoothing intensity are set, where the first smoothing window length is less than a smoothing window length threshold and the first smoothing intensity is less than a smoothing intensity threshold. The motion information is smoothed based on the length of the first smoothing window and the first smoothing intensity to preserve motion details; When the motion trajectory changes smoothly, a second smoothing window length and a second smoothing intensity are set. The second smoothing window length is greater than the smoothing window length threshold, and the second smoothing intensity is greater than the smoothing intensity threshold. Motion information is smoothed based on the second smoothing window length and the second smoothing intensity to eliminate random jitter.

[0090] In some embodiments, the changes in motion trajectories may exhibit either dramatic or stable states. If a single smoothing parameter is used, excessive smoothing may lead to the loss of important motion details when the motion trajectory changes dramatically; conversely, insufficient smoothing may fail to effectively eliminate random jitter when the motion trajectory changes smoothly, thus affecting the accuracy of subsequent behavioral pattern analysis. Therefore, a smoothing window length threshold and a smoothing intensity threshold can be determined based on the local characteristics of the motion information. The local characteristics of the motion information refer to the statistical properties exhibited by the motion data within a specific time window, such as the variance of velocity and acceleration, frequency components, or rate of change. These characteristics are used to determine whether the motion trajectory is dramatic or stable. The smoothing window length threshold and the smoothing intensity threshold are set benchmark values ​​used to distinguish the application scenarios of different smoothing strategies. The smoothing window length refers to the number of motion data points used to calculate the average or weighted average during smoothing, which determines the range of smoothing; the smoothing intensity refers to the degree of influence of the smoothing algorithm on the original data, such as the attenuation coefficient of the filter or the weight distribution, which determines the strength of the smoothing.

[0091] When motion trajectories change drastically, such as when a user performs rapid movement, jumps, or sudden turns, a first smoothing window length and a first smoothing intensity are set. The first smoothing window length is less than a smoothing window length threshold, and the first smoothing intensity is less than a smoothing intensity threshold. Based on these settings, motion information is smoothed to preserve motion details. A smaller smoothing window length and lower smoothing intensity mean less aggressive smoothing, aiming to maximize the preservation of rapid changes in motion and fine details crucial for identifying complex or dynamic behavioral patterns.

[0092] When the motion trajectory changes smoothly, such as when the user is stationary, walking slowly, or sitting, a second smoothing window length and a second smoothing intensity are set. The second smoothing window length is greater than a smoothing window length threshold, and the second smoothing intensity is greater than a smoothing intensity threshold. Based on the second smoothing window length and the second smoothing intensity, the motion information is smoothed to eliminate random jitter. A larger smoothing window length and a higher smoothing intensity mean a more aggressive smoothing process, more effectively filtering out minute random fluctuations and environmental noise, thus providing a cleaner and more stable motion signal.

[0093] To illustrate this technical solution more clearly, a specific example is used below. Assume motion data is being collected from a user. When a user engages in vigorous activity such as running or jumping, the system determines that the motion trajectory changes drastically based on local features such as speed and acceleration. In this case, the smoothing window length is set to 5 data points, and the smoothing intensity is set to 0.3 (assuming a smoothing window length threshold of 10 data points and a smoothing intensity threshold of 0.5), to ensure that key motion details such as gait characteristics during running and the instantaneous take-off and landing during jumping are preserved. When the user is sitting still or walking slowly, the system determines that the motion trajectory changes smoothly. In this case, the smoothing window length is set to 20 data points, and the smoothing intensity is set to 0.8, to effectively filter out random jitter caused by body tremors or sensor noise, thereby obtaining stable and clear posture data.

[0094] Through the above technical solution, this embodiment can intelligently adjust the smoothing parameters according to the actual changes in the motion trajectory, thereby effectively eliminating random jitter while preserving the effective details of the motion to the maximum extent. This avoids the problems of information loss when processing violent motion or insufficient noise reduction when processing steady motion in traditional fixed-parameter smoothing methods, significantly improving the accuracy and adaptability of motion information smoothing processing. As a result, it provides higher-quality input data for subsequent behavioral pattern analysis, thereby improving the accuracy of identifying functional segments from non-textual data, making the construction of life stories more refined and realistic.

[0095] The beneficial effects of implementing the embodiments of the present invention include: First, historical user data is acquired, and event clustering is performed on the historical user data to obtain event units. Then, feature extraction is performed on the event units to obtain core elements and sentiment tendencies, and semantic anchors of the event units are generated. Then, correlation analysis is performed on the current user data and semantic anchors to obtain correlation analysis results. If the correlation analysis results show that there is a correlation, then based on the sentiment tendencies and current user data, sentiment changes are identified. Finally, the importance weight of the event unit in the life story is adjusted, and the presentation content of the event unit is updated based on the current user data and importance weights. Thus, it is possible to extract events with data value by analyzing sentiment changes, so as to realize life course data analysis and improve the data value extraction capability.

[0096] like Figure 2 As shown, this embodiment of the invention also provides a life course data analysis system, including: Data acquisition module 601 is used to acquire historical user data; The event clustering module 602 is used to cluster historical user data into event units. The feature extraction module 603 is used to extract features from event units to obtain core elements and sentiment tendencies; The semantic anchor generation module 604 is used to generate semantic anchors for event units based on core elements and sentiment tendencies. The semantic anchors can be represented by keyword combinations and embedding vectors. The correlation analysis module 605 is used to perform correlation analysis on the current user data and semantic anchors to obtain the correlation analysis results; The sentiment analysis module 606 is used to identify sentiment changes based on sentiment tendencies and current user data if the correlation analysis results indicate a correlation. The weighting adjustment module 607 is used to adjust the importance weight of event units in the life story based on emotional changes. Story update module 608 is used to update the presentation content of event units based on current user data and importance weight.

[0097] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0098] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

Claims

1. A method for analyzing life course data, characterized in that, Includes the following steps: Obtain historical user data; The historical user data is clustered into event units. Feature extraction is performed on the event units to obtain core elements and emotional tendencies; Based on the core elements and the emotional tendency, semantic anchors for the event units are generated, and the representation of the semantic anchors includes keyword combinations and embedding vectors; Perform a correlation analysis on the current user data and the semantic anchor points to obtain the correlation analysis results; If the correlation analysis result indicates a correlation, then based on the sentiment tendency and the current user data, the sentiment change is identified; Adjust the importance weight of the event unit in the life story based on the emotional changes; The presentation content of the event unit is updated based on the current user data and the importance weight.

2. The method according to claim 1, characterized in that, The step of identifying emotional changes based on the emotional tendency and the current user data includes: The current user data is deconstructed to identify functional segments, which are used to represent factual restatements, causal attributions, and impact assessments of the event unit. Perceptual information is obtained by performing sensory extraction on the aforementioned functional segments; Based on the historical user data, establish a personalized user experience expression pattern; The sensory information is calibrated according to the sensory expression pattern. Based on the functional segments, logical relationships, and calibrated perceptual information, a semantic structure vector is constructed; The emotional changes are identified based on the semantic structure vector and the emotional tendency.

3. The method according to claim 2, characterized in that, The step of deconstructing the current user data and identifying functional segments includes: Identify functional expressions in the current user data, including metaphorical expressions, emotional words, or scene descriptions; If the functional expression is a metaphorical expression, then the functional fragment is identified based on the historical user data; If the functional expression is an emotional word, then the functional segment is identified based on the context of the emotional word, and the context is used to determine whether the emotional word directly expresses feelings. If the functional expression is a scene description, then the event background, cause of occurrence and subsequent impact contained in the scene description are analyzed to identify the functional segment.

4. The method according to claim 3, characterized in that, The step of identifying the functional segment based on the historical user data includes: Identify the vehicle of the metaphorical expression; Determine the core semantic meaning of the metaphor in the current context; Based on the historical user data, identify metaphorical semantic mapping rules; Based on the core semantic orientation and the metaphorical semantic mapping rules, identify the semantic association between the metaphor and the event ontology; The confidence level of the semantic association is evaluated to obtain the confidence evaluation result; Based on the confidence assessment results, the facts referred to by the metaphorical expression are determined, and the functional fragment is obtained.

5. The method according to claim 2, characterized in that, When the current user data contains non-textual data, the step of deconstructing the current user data and identifying functional segments includes: Multimodal information extraction is performed on the non-textual data to obtain visual, auditory, and motion information; The visual information is analyzed to obtain scene content, which includes objects, people, actions, and environment. The auditory information is subjected to sound event recognition to obtain sound events, which include speech, music and ambient sound; The motion information is analyzed to obtain behavioral patterns, which include walking, running, jumping, sitting, and waving. The functional segments are identified based on the scene content, the sound events, the behavioral patterns, and the semantic anchors.

6. The method according to claim 5, characterized in that, The step of performing scene content analysis on the visual information to obtain scene content includes: The visual information is preprocessed to remove noise or enhance image features; Multi-scale analysis is performed on the preprocessed visual information to identify the features of visual elements at different scales; Spatial relationship analysis is performed on the visual element features to determine the relative positions and occlusion relationships between visual elements, and multi-scale analysis results are obtained. Based on the multi-scale analysis results, the occluded or incomplete visual elements are inferred to complete the inference results. Based on the multi-scale analysis results and the completion inference results, the scene content is identified.

7. The method according to claim 5, characterized in that, The step of performing sound event recognition on the auditory information to obtain sound events includes: The auditory information is converted into the time-frequency domain to obtain a time-frequency representation of the auditory information; Noise suppression and reverberation cancellation are performed on the time-frequency representation of the auditory information to obtain a clean time-frequency representation; The pure time-frequency representation is subjected to source separation to obtain the source components; The acoustic features are obtained by extracting features from the sound source components. Based on the acoustic characteristics, the sound source components are classified to obtain the sound events.

8. The method according to claim 5, characterized in that, The step of performing behavioral pattern analysis on the motion information to obtain behavioral patterns includes: The motion information is smoothed to eliminate random jitter in the motion trajectory; Outlier removal is performed on the smoothed motion information to remove data points that do not conform to normal motion patterns; Noise filtering is applied to the motion information after outlier removal to reduce the impact of environmental interference on posture change recognition. Motion features are extracted from the noise-filtered motion information, including velocity, acceleration, angular velocity, and joint angles. Based on the motion characteristics, the motion information is classified to obtain the behavior pattern.

9. The method according to claim 8, characterized in that, The smoothing process for the motion information includes: Based on the local features of the motion information, determine the smoothing window length threshold and the smoothing intensity threshold; When the motion trajectory changes drastically, a first smoothing window length and a first smoothing intensity are set, wherein the first smoothing window length is less than the smoothing window length threshold and the first smoothing intensity is less than the smoothing intensity threshold. The motion information is smoothed based on the length of the first smoothing window and the first smoothing intensity to preserve motion details; When the motion trajectory changes smoothly, a second smoothing window length and a second smoothing intensity are set, wherein the second smoothing window length is greater than the smoothing window length threshold and the second smoothing intensity is greater than the smoothing intensity threshold. The motion information is smoothed according to the second smoothing window length and the second smoothing intensity to eliminate random jitter.

10. A life course data analysis system, characterized in that, include: The data acquisition module is used to acquire historical user data; The event clustering module is used to cluster the historical user data into event units. The feature extraction module is used to extract features from the event unit to obtain core elements and emotional tendencies; The semantic anchor generation module is used to generate semantic anchors for the event unit based on the core elements and the sentiment tendency. The semantic anchors are represented by keyword combinations and embedding vectors. The correlation analysis module is used to perform correlation analysis on the current user data and the semantic anchor points to obtain the correlation analysis results; The sentiment analysis module is used to identify sentiment changes based on the sentiment tendency and the current user data if the correlation analysis result indicates a correlation. The weighting adjustment module is used to adjust the importance weight of the event unit in the life story based on the emotional changes. The story update module is used to update the presentation content of the event unit based on the current user data and the importance weight.

Images