Data interaction method and device, electronic equipment and computer readable storage medium

Abstract

The invention provides a data interaction method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of artificial intelligence and the like. According to the specific implementation scheme, natural language data of a user is received; performing feature separation on the natural language data to generate a discrete fact data packet with non-overlapped main features and a continuous scene vector; updating a preset fact database based on the discrete fact data packet to obtain an updated fact database; carrying out attenuation processing on the continuous scene vector, and writing the attenuated scene vector into a preset time sequence database; calculating dynamic view parameters based on the current scene vector in the time sequence database; and based on the dynamic view parameters, performing runtime reconstruction on the data in the updated fact database, and generating and outputting response data.

Description

Technical Field

[0001] This disclosure belongs to the field of computer technology, specifically relating to the technical fields of natural language processing, large language models, deep learning, etc., and in particular to a data interaction method and apparatus, electronic devices, and computer-readable storage media. Background Technology

[0002] With the rapid development of large language models, artificial intelligence is evolving from simple chatbots to autonomous intelligent agents. In the next generation of computer system architecture (AI-Native OS), intelligent agent systems are expected to serve as the core kernel, responsible for understanding user intent, breaking down complex tasks, and scheduling system resources.

[0003] Current intelligent agent systems primarily rely on natural language as the core interaction and storage medium. The system maintains its memory of tasks by keeping a constantly growing "context window" or "dialogue history".

[0004] Although the length limit of the context window is being continuously broken, when dealing with long-term tasks, architectures that rely solely on natural language streams face the bottleneck of increased information entropy at the physical level. As a result, this natural language stream processing method results in a poor dialogue experience between users and intelligent agent systems. Summary of the Invention

[0005] This disclosure provides a data interaction method and apparatus, an electronic device, and a computer-readable storage medium.

[0006] According to the first aspect, a data interaction method is provided, which includes: receiving natural language data from a user; performing feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; updating a pre-set fact database based on the discrete fact data packets to obtain an updated fact database; writing the continuous scenario vectors into a pre-set time-series database, and performing attenuation processing on the continuous scenario vectors based on the current system time and the timestamps of historical scenario vectors in the time-series database to obtain a current scenario vector; calculating dynamic view parameters based on the current scenario vectors; and performing runtime reconstruction on the data in the updated fact database based on the dynamic view parameters to generate and output response data.

[0007] According to a second aspect, a data interaction device is provided, comprising: a receiving unit configured to receive natural language data from a user; a separation unit configured to perform feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; an updating unit configured to update a pre-set fact database based on the discrete fact data packets to obtain an updated fact database; an attenuation unit configured to write the continuous scenario vectors into a pre-set time-series database and perform attenuation processing on the continuous scenario vectors based on the current system time and the timestamps of historical scenario vectors in the time-series database to obtain a current scenario vector; a calculation unit configured to calculate dynamic view parameters based on the current scenario vectors; and a reconstruction unit configured to perform runtime reconstruction of the data in the updated fact database based on the dynamic view parameters to generate and output response data.

[0008] According to a third aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first aspect.

[0009] According to a fourth aspect, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to perform the method described in any implementation of the first aspect.

[0010] The data interaction method and apparatus provided in the embodiments of this disclosure first receive natural language data from a user; second, perform feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; third, update a pre-set fact database based on the discrete fact data packets to obtain an updated fact database; next, write the continuous scenario vectors into a pre-set time-series database, and perform attenuation processing on the continuous scenario vectors based on the timestamps of historical scenario vectors in the time-series database and the current system time to obtain the current scenario vector; then, calculate dynamic view parameters based on the current scenario vectors; finally, reconstruct the data in the updated fact database at runtime based on the dynamic view parameters to generate and output response data. This data interaction method breaks the chaotic state of natural language and is no longer limited to the natural language processing level. Through hierarchical state compression, it transforms the irreversible information entropy increase process into an ordered state machine flow, which can rigidly lock facts like a traditional database, effectively preventing information drift, and can perceive scenarios like a human assistant, realizing dynamic empathy. This enables the system corresponding to the data interaction method to operate stably for a long time in complex environments.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This is a flowchart of one embodiment of the data interaction method according to this disclosure;

[0014] Figure 2 This is a schematic diagram of the system framework corresponding to the publicly disclosed data interaction method;

[0015] Figure 3 This is a schematic diagram of the structure of one embodiment of the data interaction device disclosed herein;

[0016] Figure 4 This is a block diagram of an electronic device used to implement the data interaction method of the embodiments of this disclosure. Detailed Implementation

[0017] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0018] The technical solutions of this disclosure are illustrated below through specific embodiments. It should be understood that one or more steps mentioned in this disclosure do not preclude the existence of other methods and steps before or after the combined steps, or that other methods and steps may be inserted between these explicitly mentioned steps. It should also be understood that these examples are for illustrative purposes only and are not intended to limit the scope of this disclosure. Unless otherwise stated, the numbering of each method step is only for the purpose of identifying each method step, and not to limit the order of each method or to limit the scope of implementation of this disclosure. Changes or adjustments to their relative relationships, without substantial changes to the technical content, can also be considered as within the scope of implementation of this disclosure.

[0019] The raw materials and instruments used in the examples are not subject to any specific restrictions on their source; they can be purchased from the market or prepared according to conventional methods known to those skilled in the art.

[0020] Traditional robots, when interacting with information, rely solely on natural language streams, which face a bottleneck of increasing information entropy. This bottleneck manifests in the following three major technical challenges:

[0021] Semantic drift caused by information entropy increase: Natural language is essentially an unstructured, highly redundant, and ambiguous data format. As the number of dialogue rounds increases, the noise (invalid information) in the context grows exponentially, while the signal density of core facts (such as task deadlines and budget constraints) is diluted. The attention mechanism of large language models can "get lost in the middle" when processing long texts, easily "forgetting" the initially set rigid constraints after multiple rounds of interaction, resulting in a generated plan that deviates from the initial goal. For example, if a user sets "no later than Friday" in the first round, the intelligent agent system may schedule a task for Saturday in the 20th round due to idle chatter.

[0022] Lacking dynamic scheduling capabilities with state awareness, existing intelligent agent architectures are typically statically responsive. They can only perform semantic understanding of text, but cannot quantify the user's "psychological state" into system operating parameters, nor can they distinguish between emotional outbursts and rational commands. When users are in a state of "high anxiety and high cognitive load," existing AI still mechanically outputs a large number of complex task lists based on prompts. This "correct but useless" output further increases the user's cognitive burden, causing the user to abandon execution and ultimately leading to system task failure.

[0023] The dilemma of context compression: To address the token limitation in intelligent agent systems, existing technologies typically employ two approaches: storing historical slices and summarizing history. However, both have drawbacks. Storing historical slices loses temporal continuity and a sense of global state, resulting in retrieved fragments often lacking current context. Summarizing history is a lossy compression method, often retaining explicit facts (what was done) while discarding implicit states (the user's emotions, tone, and subtext). These implicit states are crucial for maintaining long-term trust relationships.

[0024] This disclosure provides a data interaction method. Figure 1 A flow 100 of an embodiment of the data interaction method according to this disclosure is shown, the data interaction method including the following steps:

[0025] Step 101: Receive the user's natural language data.

[0026] In this embodiment, the natural language data can be text data obtained after processing the multimodal interactive materials input by the user. Specifically, the interactive materials include at least one of interactive voice, interactive image, and interactive text. Step 101 includes: determining the type of multimodal data in the interactive materials; performing corresponding material-to-text processing on the interactive materials based on the type to obtain the user's natural language data.

[0027] Optionally, step 101 above includes: receiving text input information from the user or a text stream transcribed from the user's speech. The text input information or text stream is then encapsulated into standardized natural language data. This natural language data can be an EventFrame object, which includes a timestamp, a user ID, and the original text data of the corresponding text input information or text stream.

[0028] Step 102: Perform feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features.

[0029] In this embodiment, the discrete fact data package includes objective fact data and constraints, while the continuous scenario vector includes subjective scenario features.

[0030] In this embodiment, feature separation is the process of forcibly decomposing an unstructured natural language data stream into two mathematically independent subspaces (i.e., the main features do not overlap and do not interfere with each other).

[0031] In this embodiment, the discrete fact data package refers to a structured data set generated after separation, used to map the constraints and objectives of the objective physical world. The discrete fact data package includes objective fact data and constraint conditions.

[0032] In this embodiment, the continuous scenario vector is a low-dimensional Euclidean space vector generated after separation, used to map the user's subjective psychological state or event scenario state. The continuous scenario vector includes subjective scenario features, such as emotional valence features and arousal features. The specific value of the feature is called the feature value, such as the value of the emotional valence feature is called the emotional valence feature value. The discrete fact data package and the continuous scenario vector are information extracted from natural language data. The main features of the discrete fact data package and the continuous scenario vector are substantially independent, and the main features of the two do not overlap.

[0033] In this embodiment, the input natural language statement is parsed in two parallel paths: one path uses dependency-semantic role labeling to extract atomizable propositions (entities, attributes, events, time, negation / modality constraints, etc.) into key-value-predicate triples, which are then packaged into discrete fact data packets after entity linking and conflict resolution; the other path uses a pre-trained sentiment-tone-intention joint model to map the remaining non-atomic subjective colors (emotional polarity, speaker's perspective, occasion formality, metaphor intensity, etc.) to a 128-dimensional continuous latent space to obtain a continuous context vector; the two output orthogonal paths are controlled so that the discrete fact data packets and the continuous context vector are strictly zero-correlated in the sense of vector dot product, thereby achieving non-overlapping representations of objective facts and subjective contexts in the same semantic space.

[0034] Step 103: Based on the discrete fact data packet, update the pre-set fact database to obtain the updated fact database.

[0035] In this embodiment, the fact database is a database used to store discrete fact data packets at various times.

[0036] In this embodiment, to update the fact database, discrete fact data packets are first parsed to extract entities, attributes, and relationships. Then, the corresponding record is located in the existing database based on the primary key or unique identifier. If it already exists, a covering update or field-level merging is performed; otherwise, an insert is performed. During the update process, indexes, foreign key constraints, and triggers are maintained simultaneously to ensure consistency and integrity. Finally, a timestamp and source marker are written to complete an atomic transaction commit, enabling the database to reflect the latest set of facts instantly. This implementation is based on graph-based event appending and state merging, which can preserve the complete historical event chain, rather than simply overwriting fields.

[0037] Step 104: Write the continuous scenario vector into a pre-set time series database, and perform attenuation processing on the continuous scenario vector based on the timestamps of the current system time and the historical scenario vectors in the time series database to obtain the current scenario vector.

[0038] In this embodiment, attenuation processing is a data processing method that simulates the biological memory forgetting mechanism. Specifically, based on the difference between the current system time and the timestamp of the historical context vector, the intensity of the continuous context vector is exponentially weighted and downgraded. Here, the historical context vector is the context vector in the time series database that corresponds to the continuous context vector and belongs to the previous moment.

[0039] In this embodiment, step 104 includes: querying the most recently written historical scenario vector and its corresponding timestamp from the time series database; calculating the time difference between the current system time and the timestamp; and calculating the current scenario vector using a recursive weighted formula. As shown in equation (1), the current scenario vector includes both the current instantaneous state and the decayed historical afterimage.

[0040] (1)

[0041] In equation (1), It is a continuous scenario vector; The scenario vector stored in the previous time step is obtained from the current scenario vector in the previous time step; For the time difference to be calculated; Update the weights for the continuous scenario vector (e.g., 0.6, indicating that newly occurring events account for 60% of the weight). Calculate the current scenario vector. The current timestamp is written as a new data point to the time series database.

[0042] Step 105: Calculate the dynamic view parameters based on the current scenario vector.

[0043] In this embodiment, the dynamic view parameters are a set of control variables calculated based on the current scenario vector, used to control the data display format and interaction strategy.

[0044] In this embodiment, based on the current scenario vector of the time series database, the full-dimensional values ​​of the latest sampling time are first extracted as the benchmark. Then, the first-order difference of each dimension is performed within the sliding time window to obtain the instantaneous rate of change. The benchmark value and the rate of change are then fused into a dynamic weight vector according to preset weights. Next, the original dimensions are weighted and dimensionality reduced using this vector. Principal component analysis is used to retain principal components whose cumulative contribution rate is greater than or equal to the contribution threshold (e.g., 85%). Their projection coefficients are normalized and output as dynamic view parameters, realizing a visualization perspective that adaptively adjusts with data evolution.

[0045] Step 106: Based on the dynamic view parameters, perform runtime reconstruction of the data in the updated fact database to generate and output response data.

[0046] In this embodiment, response data is the reply data to the natural language data input by the user. Runtime reconstruction refers to the process of real-time cropping, filtering and reorganizing of the original data in the static database based on dynamic view parameters during the data reading phase.

[0047] In this embodiment, during runtime, the system reconstructs the original data in the updated fact database in real time based on the input dynamic view parameters (such as field filtering, aggregation dimensions, sorting rules, etc.): first, it filters valid records according to the parameters, then it groups, calculates and maps according to the view definition, and finally assembles and outputs a response data structure that meets the front-end requirements.

[0048] The data interaction method provided in the embodiments of this disclosure first receives natural language data from a user; second, it performs feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; third, it updates a pre-set fact database based on the discrete fact data packets to obtain an updated fact database; fourth, it performs attenuation processing on the continuous scenario vectors and writes the attenuated scenario vectors into a pre-set time-series database; fifth, it calculates dynamic view parameters based on the current scenario vectors in the time-series database; and finally, it performs runtime reconstruction on the data in the updated fact database based on the dynamic view parameters to generate and output response data.

[0049] In some embodiments of this disclosure, the above-mentioned calculation of dynamic view parameters based on the current scenario vector includes: performing a weighted moving average on the current scenario vector to obtain a system weighted state value; and obtaining dynamic view parameters based on the system weighted state value through an anti-entropy adjustment strategy.

[0050] In this optional implementation, the weighted moving average is a statistical processing method used to analyze the time series of data points. Unlike the simple moving average (where all data points have the same weight), the weighted moving average in this disclosure assigns different weights to data points at different times. Generally, data points closer to the current time have higher weights, while data points farther away have lower weights. The system weighted state value input obtained through the weighted moving average acts as a low-pass filter, filtering out occasional emotional noise, making the system's state judgment smoother and more stable, and preventing the system from suddenly switching to "dimensionality reduction protection mode" due to a user's unintentional remark.

[0051] In this optional implementation, the dynamic view parameters obtained through an anti-entropy adjustment strategy based on the system's weighted state values ​​include: setting a target value representing the user's cognitive load level; calculating the deviation between the system's weighted state values ​​and the target value in real time; and calculating the anti-entropy control quantity based on the deviation using a PID (proportional-integral-derivative) algorithm, and linearly mapping the calculated anti-entropy control quantity to the dynamic view parameters. In the PID algorithm, the proportional term responds to the magnitude of the current deviation; the integral term eliminates long-standing steady-state errors; and the derivative term predicts future trends. The larger the error (higher the anxiety) represented by the anti-entropy control quantity mapped to the dynamic view parameters, the fewer entries are displayed in the dynamic view parameters, and the changes are continuous and smooth, rather than abrupt.

[0052] In the optional implementation, the N most recently written scenario vectors in the time series database are first weighted and moved averaged according to the time decay coefficient to obtain a smoothed system weighted state value vector. Then, the mean vector is input into the anti-entropy adjustment module: the KL divergence between the mean vector and the dynamic view parameters of the previous period is calculated as an entropy increase index. If the divergence exceeds the threshold, the mean vector is compensated and corrected according to the negative feedback ratio. Finally, the dynamic view parameters with both smoothness and low latency are output, which improves the reliability of the obtained dynamic view parameters and enables the dynamic view parameters to be effectively used for downstream visualization or decision-making.

[0053] Optionally, the above-mentioned dynamic view parameters obtained based on the system weighted state values ​​through an anti-entropy adjustment strategy include: mapping the components of each vector in the system weighted state values ​​to a fuzzy set, performing fuzzy rule reasoning on the fuzzy set using a preset fuzzy rule base to obtain the reasoning result; and using the centroid method to convert the reasoning result into dynamic view parameters.

[0054] Optionally, the dynamic view parameters include: an information capacity limit value. The calculation of the dynamic view parameters based on the current scenario vector in the time-series database includes: extracting the cognitive load index from the current scenario vector; substituting the cognitive load index into a preset hyperbolic suppression function for calculation to obtain the information capacity limit value in the dynamic view parameters; wherein, the hyperbolic suppression function is configured such that the information capacity limit value and the cognitive load index have a non-linear inverse relationship, and when the cognitive load index reaches a preset saturation value, the information capacity limit value converges to a preset minimum positive integer.

[0055] To achieve a smooth transition, this invention introduces a hyperbolic suppression function as shown in equation (2) to fine-tune the parameters. The hyperbolic suppression function is used to define the information capacity control equation. In equation (2), the information capacity limit value is... Cognitive load metrics of users They are inversely proportional.

[0056] (2)

[0057] In formula (2): The baseline capacity is 10. For the current cognitive load ( ), This is the sensitivity coefficient (empirical value taken as 5.0). Indicates rounding down, the mathematical derivation of equation (2) is: when hour, ;when (High voltage) This mathematically guarantees that the system under high pressure will never output a long list.

[0058] In some embodiments of this disclosure, the above-mentioned method of obtaining dynamic view parameters based on system weighted state values ​​through an anti-entropy adjustment strategy includes: acquiring preset on thresholds and off thresholds, wherein the on threshold is greater than the off threshold; comparing the system weighted state value with the on threshold and the off threshold respectively; determining the current system operating mode as the dimensionality reduction protection mode in response to the system weighted state value being greater than the on threshold, or in response to the previous system operating mode being the dimensionality reduction protection mode and the system weighted state value being greater than the off threshold; determining the current system operating mode as the full information mode in response to the system weighted state value being less than or equal to the off threshold; and retrieving the corresponding set of control parameters from a preset parameter mapping relationship based on the determined system operating mode as dynamic view parameters; wherein the set of control parameters corresponding to the dimensionality reduction protection mode represents the compressed information capacity, and the set of control parameters corresponding to the full information mode represents the full information capacity.

[0059] In this optional implementation, the anti-entropy adjustment strategy treats the human-computer interaction process as a closed-loop feedback control problem, aiming to maintain the user's information reception entropy within the optimal range. To this end, the execution entity running on the data processing method defines an adjustable policy parameter space and maps the scenario vectors in the time-series database to this policy parameter space through a set of explicit control laws. In the policy parameter space, the system's (e.g., an intelligent agent system) behavior is quantified into four independently adjustable scalar parameters, which constitute the output of the control law, defining dynamic view parameters. The definitions of each component in the dynamic view parameters are shown in Table 1.

[0060] Table 1

[0061]

[0062] In this optional implementation, the parameter mapping relationship can be represented by a control law lookup table, which is shown in Table 2.

[0063] Table 2

[0064]

[0065] In this optional implementation, a preset activation threshold (greater than the deactivation threshold) is first taken, and the system weighted state value of the scenario vector is compared with these two thresholds. If the mean is higher than the activation threshold, the system immediately switches to the dimensionality reduction protection mode and outputs dynamic view parameters for compressing information capacity. If the system is already in the dimensionality reduction protection mode and the mean has fallen back to the deactivation threshold or lower, the system exits the mode and outputs dynamic view parameters for restoring full information capacity. If the mean is between the two thresholds, the existing mode and corresponding parameters remain unchanged.

[0066] In some embodiments of this disclosure, the above-mentioned method of updating a pre-set fact database based on discrete fact data packets to obtain an updated fact database includes: extracting task description text from the discrete fact data packets and converting the task description text into a high-dimensional semantic vector; calculating the cosine similarity between the high-dimensional semantic vector and the task vectors already stored in the fact database; in response to the cosine similarity being greater than a preset semantic overlap threshold, updating the fields of the corresponding existing task records in the fact database based on the constraints in the discrete fact data packets to obtain an updated fact database; and in response to the cosine similarity being less than the semantic overlap threshold, writing the discrete fact data packets as new records into the fact database to obtain an updated fact database.

[0067] In this optional implementation, a high-dimensional semantic vector is a semantic vector that reaches a certain dimension, such as a semantic vector with 512 dimensions or higher. Fact databases (such as...) Figure 2It is not just a storage device; it is a hybrid database with semantic deduplication capabilities. It solves the problem of traditional chatbots repeatedly recording the same task in multi-turn dialogues. The structured storage of the fact database is used to store precise fields such as Task_ID, Status, Deadline, and Priority; the unstructured index of the fact database uses the HNSW algorithm to store the embedding vector of Task_Description for similarity retrieval. When discrete fact data packets enter, the execution entity running on them does not directly insert them, but first calculates the semantic distance as shown in Equation (3) through high-dimensional semantic vectors to determine whether it is "added" or "modified". Let the existing task set in the database be in Equation (3). The newly entered task description text is Define the similarity function. Set threshold (e.g., 0.85).

[0068] (3)

[0069] In a specific example, the fact database already contains: {id: 101, desc: "Write a weekly report"}, and the user-input task description text is: "Change the report time to Friday" -> recognized as t new If the semantic distance is calculated as: Sim(“Write Weekly Report”, “Report”) > 0.85, then the operation is: triggering an SQL update to update the field of UPDATE tasks SET due_date=FridayWHERE id=101, instead of inserting a new record.

[0070] In this optional implementation, the task description text is first extracted from the discrete fact data package and mapped to a 512-dimensional high-dimensional semantic vector using a pre-trained language model. Then, the cosine similarity between this high-dimensional semantic vector and the existing task vectors in the fact database is calculated. If the maximum similarity is higher than the preset semantic overlap threshold (which can be set according to requirements, such as 0.85), it is determined to be the same task. Then, the latest constraints in the data package are used to perform incremental updates on the fields of the corresponding records in the database. Otherwise, the entire data package is appended to the database table as a new task record, thus completing the online expansion and dynamic maintenance of the fact database.

[0071] Optionally, the above-mentioned method of updating a pre-set fact database based on discrete fact data packets to obtain an updated fact database further includes: setting a high confidence threshold ( (e.g., 0.95) and low confidence threshold ( (e.g., 0.75). If the calculated cosine similarity is within [...]. , If the specified interval (i.e., the ambiguity region) is reached, the system suspends the current database write operation. The system then bases its current intent clarity on the context vector corresponding to the discrete fact data packets in the time-series database. Generate a confirmation request. If If the user is highly alert (in a state of high alert), a direct question will be generated: "Do you mean you want to modify the [old task title]?" For users with low user confusion, a guided query is generated: "I found a similar [old task], do you want to merge it?" The system receives the user's response (Yes / No) to this confirmation request and only performs an "update" operation upon receiving a positive response; otherwise, it performs an "insert" operation. This solves the problem of data overwriting or duplicate task creation caused by ambiguous user instructions (such as "change that one"), ensuring the purity of the fact database.

[0072] In some embodiments of this disclosure, the above-mentioned attenuation processing of continuous scenario vectors based on the timestamps of historical scenario vectors in the time-series database to obtain the current scenario vector includes: calculating the time difference based on the timestamps of the current system time and historical scenario vectors in the time-series database; using the time difference, weighting the emotional valence feature value and arousal feature value in the continuous scenario vector using an exponential decay function to obtain the attenuated feature value; combining the attenuated feature values ​​into the current scenario vector, and writing the current scenario vector and the current system time into the time-series database; wherein, the decay coefficient of the exponential decay function is inversely proportional to a preset forgetting time constant, so that the attenuated scenario vector numerically converges to a preset neutral baseline state.

[0073] In this optional implementation, continuous scenario vectors are used to map the user's subjective psychological state. The scenario subspace L2 containing the continuous scenario vectors is a continuous, low-dimensional Euclidean vector space. To simulate the transient nature of human memory and emotions, the L2 state space follows an exponential decay model. This ensures that historical emotional fluctuations will not permanently interfere with the long-term operation of the system. It is Dimensional column vector. In a preferred embodiment of this disclosure, The vector is defined as shown in equation (4):

[0074] (4)

[0075] The physical meaning and value range of each dimension in equation (4) are as follows: Emotional valence metric, with values ​​ranging from -1.0 to 1.0. -1.0 represents extreme negativity, and 1.0 represents extreme positivity. : Emotional arousal characteristic, with values ​​[0.0, 1.0]. 0 represents drowsiness / calmness, and 1 represents alertness / anger; Cognitive load, with values ​​[0.0, 1.0], is calculated based on the perplexity and syntactic complexity of the input text; Clarity of Intent, with values ​​[0.0, 1.0]. 1 represents a clear instruction, and 0 represents vague or rambling instructions.

[0076] In this optional implementation, the timestamp of the historical context vector is the time when the state of the historical context vector was generated. The execution entity running on it will adjust the time according to the current system time. State generation time of historical context vector The difference (at the last interaction time), i.e., the time difference, attenuates the intensity of the continuous scenario vector. Let... For the current system time, The last interaction moment, Therefore, the update of the state of the scenario vector in the time series database depends not only on the input, but also on the passage of time, as shown in equation (5):

[0077] (5)

[0078] In equation (5): Forgetting time constant , The attenuation coefficient is calculated using equation (5) in the code through lazy calculation or a scheduled task. When the scheduler reads the time-series database, it first applies this formula to weight the stored historical values.

[0079] In a specific example, suppose The user was extremely angry at 9:00 AM. If the user does not return until 2:00 PM ( Upon returning, the system read the remaining anger value as follows: The executing entity will automatically determine that the user has returned from "extreme irrationality" to "moderate emotion", thereby automatically deactivating the most stringent "dimensional reduction protection mode".

[0080] Optionally, the above data processing method further includes: performing statistical analysis on the scenario vectors in the time-series database at a preset period (e.g., daily / weekly) to calculate the baseline emotional fingerprint. The average cognitive load baseline for the user during a specific time period (e.g., Monday morning) is calculated. Based on this baseline emotional fingerprint, the baseline capacity parameter of the adaptive scheduler is dynamically calibrated. If the user's average cognitive load baseline is consistently high (e.g., chronic anxiety user), the system automatically lowers the default baseline capacity parameter (e.g., from 10 to 5) to achieve personalized "default dimensionality reduction." This enables the system to possess "meta-learning" capabilities, adjusting interaction strategies not only based on current emotional responses but also on the user's long-term personality traits.

[0081] Optionally, the above data processing method further includes: monitoring change events in the fact database. If a task status marked as "high priority" in the fact database changes from Pending to Complete (i.e., the user has completed a critical task), a reward reset signal is triggered in the situation vector state machine of the time-series database. This forcibly multiplies the current cognitive load and negative emotional valence feature values ​​by a reset coefficient (e.g., 0.5), generating and writing a new situation vector that instantly reduces stress. This simulates the "relief" feeling humans experience after completing a difficult task. Compared to simple time decay, this mechanism can more realistically and logically deactivate the "dimensionality reduction protection mode," improving the user experience.

[0082] In some embodiments of this disclosure, the above-described runtime reconstruction of data in the updated fact database based on dynamic view parameters to generate and output response data includes: parsing the dynamic view parameters to obtain information capacity limit values, time horizon window values, and granularity levels; constructing time filtering conditions based on the time horizon window values ​​and selecting corresponding task-level fields based on the granularity levels; pushing the time filtering conditions and task-level fields down to the database query layer of the updated fact database; constructing a query statement including an entry number limit clause, the parameters of which are determined by the information capacity limit value; executing the query statement to extract a subset of data that matches the current context from the updated fact database as response data, and outputting the response data.

[0083] In this optional implementation, the scheduler receives dynamic view parameters and performs runtime reconstruction of the static facts in the updated fact database. This runtime reconstruction is not a simple query, but rather a trimming and reorganization of the query results.

[0084] In this optional implementation method, the following can be adopted: Figure 2The scheduler shown can map the continuous state space of a time-series database to a discrete parameter space of a scheduling policy. Internally, the scheduler maintains a lookup table (as shown in Table 3) or a decision tree to generate dynamic view parameters in the query statement. The scheduler uses the dynamic view parameters described above. Dynamically rewrite database query statements.

[0085] Table 3

[0086]

[0087] In this optional implementation, the continuous scenario vector is a feature that distinguishes this disclosure from traditional databases or chatbots. The continuous scenario vector represents the user's subjective psychological state (such as anxiety level, cognitive load, mood). The fact database stores objective facts (such as writing code, attending meetings). The scheduler must first read the scenario vector in the time series database in order to determine how to perform the task corresponding to the objective facts in the fact database.

[0088] In this optional implementation, when the scheduler reads a cognitive load > 0.8 (high-pressure state) in the scenario vector of the time-series database, it triggers the lookup table shown in Table 2 to forcibly increase the information capacity. Set to 1 (i.e., display only 1 task, entering "dimensionality reduction protection mode"). Without processing the scenario vectors mentioned above, the scheduler wouldn't know the user is currently anxious and would still mechanically output the tasks corresponding to the 10 fact database entries, leading to user frustration (increased entropy). The purpose of the scheduler processing the scenario vectors is to dynamically trim and reconstruct the "facts to be displayed" in the fact database based on the "user's current state." This is a crucial step in achieving "anti-entropy" and "crash prevention."

[0089] In this optional implementation, the dynamic view parameters are first broken down into three elements: "maximum number of records to retrieve, viewing time, and granularity level to select." Time windows are used to construct filter conditions, and the corresponding task-level fields are selected according to granularity and pushed down to the query layer of the fact database. Then, a LIMIT clause calculated from the "information capacity limit value" is added to the SQL. After execution, the retrieved data that is both fresh and well-controlled in quantity is directly output as the response data.

[0090] Optionally, the above-mentioned runtime reconstruction of the data in the updated fact database based on dynamic view parameters to generate and output response data also includes: parsing the emotional intervention weight in the dynamic view parameters. The wording of the generated text is adjusted based on the emotional intervention weight, and the UI rendering parameters of the response data output interface are adjusted synchronously. Specifically, if the emotional intervention weight is greater than the emotional threshold (the emotional threshold represents high pressure / needs reassurance, such as an emotional threshold of 0.8), the output interface mode is switched to "minimalist mode" (unnecessary buttons are hidden), and the color scheme is switched to "cool colors / low saturation" (reducing visual stimulation). If it is voice output, the speech rate and pitch are reduced. If the emotional intervention weight is less than the efficiency threshold (the efficiency threshold represents efficient / normal working state, such as an efficiency threshold of 0.3), the complete information density is displayed through the interface, and the voice output uses a standard speech rate. This embodiment extends the "anti-entropy" concept from the data layer to the presentation layer, comprehensively reducing the user's cognitive load.

[0091] In some embodiments of this disclosure, the above-described feature separation of natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features includes: inputting natural language data into a pre-trained large language model; constraining the decoding process of the large language model using predefined text data patterns to generate a first data stream and a second data stream in parallel; filtering the first data stream for stop words, removing fields containing subjective emotion words, and generating discrete fact data packets; calculating a cognitive load index based on the textual physical features of the natural language data, and combining it with the emotion classification probability in the second data stream to synthesize a continuous scenario vector; the discrete fact data packets and the continuous scenario vectors are mutually exclusive in data structure.

[0092] In this optional implementation, the original natural language text is fed into a pre-trained large language model. At the decoding end, two controlled generation paths are simultaneously mounted: one is constrained by a "fact-stop word" template, which only allows the output of objective entities such as noun phrases, numbers, and times, and filters out subjective emotion words in real time, and finally packages them into discrete fact data packets; the other is constrained by an "emotion-cognition" template, which outputs the emotion probability distribution of each word in parallel. At the same time, the original text is statistically analyzed at the front end for physical features such as character length, syntactic complexity, and the proportion of rare words, which are converted into cognitive load scores. These scores are then weighted and summed frame by frame with the emotion probability to form a multi-dimensional floating-point sequence, i.e., a continuous context vector. The two paths are independently stored on disk. The fact packets do not contain any floating-point emotion values, and the context vectors do not contain any specific entity strings, thus achieving orthogonal isolation at the data structure level.

[0093] Optionally, the above-mentioned feature separation of natural language data to generate discrete fact data packages and continuous context vectors with non-overlapping main features also includes: using a pre-defined sentiment dictionary or sentiment classification model to perform a secondary scan on the Task_Description field in the generated discrete fact data package. If the scan result shows that the field still contains high-intensity sentiment words (e.g., "annoying report"), then the orthogonal separation is deemed to have failed. The discrete fact data package containing sentiment words is then sent back to the semantic compressor, and the "negative constraints" in the prompt words are dynamically enhanced, such as: "adjectives are prohibited, only verbs and nouns are retained," and the discrete fact data package is regenerated using the enhanced prompt words. This prevents "sentiment leakage" caused by the probabilistic output of the Large Model (LLM), ensuring that the fact database stores absolutely objective facts and avoiding "garbage in, garbage out."

[0094] The following is combined Figure 2 The following is a detailed description of an embodiment of the data processing method disclosed herein: A user inputs interactive materials into the user interface, which are then processed (e.g., speech-to-text) to obtain natural language data. This natural language data is encapsulated into standardized EventFrame objects. A semantic compressor decouples the natural language data using orthogonal features, generating discrete fact data packets and continuous scenario vectors with non-overlapping main features. A structured parser parses the discrete fact data and continuous scenario vectors respectively, writing the discrete fact data into a fact database and the continuous sentiment vectors into a time-series database. A policy generator attenuates the continuous scenario vectors in the time-series database and writes the attenuated scenario vector set to the current scenario vector into the time-series database. A scheduler calculates dynamic view parameters for the current scenario vector in the time-series database and sends these dynamic view parameters to the execution agent. By reconstructing the dynamic view parameters at runtime, response data is generated and output to the user interface.

[0095] Further reference Figure 3 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a data interaction device, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0096] like Figure 3As shown, the data interaction device 300 provided in this embodiment includes: a receiving unit 301, a separation unit 302, an update unit 303, an attenuation unit 304, a calculation unit 305, and a reconstruction unit 306. The receiving unit 301 can be configured to receive natural language data from a user. The separation unit 302 can be configured to perform feature separation on the natural language data, generating discrete fact data packets and continuous scenario vectors with non-overlapping main features; wherein the discrete fact data packets include objective fact data and constraints, and the continuous scenario vectors include subjective scenario features. The update unit 303 can be configured to update a pre-set fact database based on the discrete fact data packets to obtain an updated fact database. The attenuation unit 304 can be configured to write the continuous scenario vectors into a pre-set time-series database and attenuate the continuous scenario vectors based on the current system time and the timestamps of historical scenario vectors in the time-series database to obtain the current scenario vector. The calculation unit 305 can be configured to calculate dynamic view parameters based on the current scenario vector. The aforementioned reconstruction unit 306 can be configured to perform runtime reconstruction of the updated fact database based on dynamic view parameters, and generate and output response data.

[0097] In this embodiment, the specific processing and technical effects of the receiving unit 301, separating unit 302, updating unit 303, attenuation unit 304, calculation unit 305, and reconstruction unit 306 in the data interaction device 300 can be found in the following references. Figure 1 The relevant descriptions of steps 101, 102, 103, 104, 105, and 106 in the corresponding embodiments will not be repeated here.

[0098] In some embodiments of this disclosure, the above-mentioned calculation unit 305 is configured to: perform weighted moving average processing on the current scenario vector to obtain a system weighted state value; and obtain dynamic view parameters based on the system weighted state value through an anti-entropy adjustment strategy.

[0099] In some embodiments of this disclosure, the computing unit 305 is further configured to: acquire preset enable thresholds and disable thresholds, wherein the enable threshold is greater than the disable threshold; compare the system weighted state value with the enable threshold and disable threshold respectively; in response to the system weighted state value being greater than the enable threshold, or in response to the system operating mode being dimensionality reduction protection mode and the system weighted state value being greater than the disable threshold at the previous time, determine that the system operating mode at the current time is dimensionality reduction protection mode; in response to the system weighted state value being less than or equal to the disable threshold, determine that the system operating mode at the current time is full information mode; based on the determined system operating mode, retrieve the corresponding set of control parameters from the preset parameter mapping relationship as dynamic view parameters; wherein the set of control parameters corresponding to the dimensionality reduction protection mode represents the compressed information capacity, and the set of control parameters corresponding to the full information mode represents the full information capacity.

[0100] In some embodiments of this disclosure, the update unit 303 is configured to: extract task description text from discrete fact data packets and convert the task description text into high-dimensional semantic vectors; calculate the cosine similarity between the high-dimensional semantic vectors and existing task vectors in the fact database; in response to a cosine similarity greater than a preset semantic overlap threshold, update the fields of the corresponding existing task records in the fact database based on the constraints in the discrete fact data packets to obtain an updated fact database; and in response to a cosine similarity less than the semantic overlap threshold, write the discrete fact data packets as new records into the fact database to obtain an updated fact database.

[0101] In some embodiments of this disclosure, the attenuation unit 304 is configured to: calculate the time difference based on the timestamps of the current system time and the historical scenario vectors in the time series database; based on the time difference, use an exponential decay function to perform weighted calculations on the emotional valence feature value and arousal feature value in the continuous scenario vector to obtain the attenuated feature value; combine the attenuated feature value into the current scenario vector, and write the current scenario vector and the current system time into a preset time series database; wherein, the attenuation coefficient of the exponential decay function is inversely proportional to the preset forgetting time constant, so that the attenuated scenario vector numerically converges to the preset neutral reference state.

[0102] In some embodiments of this disclosure, the reconstruction unit 306 is configured to: parse dynamic view parameters to obtain information capacity limit values, time horizon window values, and granularity levels; construct time filtering conditions based on the time horizon window values ​​and select corresponding task level fields based on the granularity levels; push the time filtering conditions and task level fields down to the database query layer of the updated fact database; construct a query statement including an entry number limit clause, the parameters of which are determined by the information capacity limit values; execute the query statement, extract a subset of data that matches the current context from the updated fact database as response data, and output the response data.

[0103] In some embodiments of this disclosure, the separation unit 302 is configured to: input natural language data into a pre-trained large language model; constrain the decoding process of the large language model using a predefined text data pattern, and generate a first data stream and a second data stream in parallel; perform stop word filtering on the first data stream, remove fields containing subjective emotion words, and generate discrete fact data packets; calculate cognitive load indicators based on the text physical features of the natural language data, and combine them with the emotion classification probabilities in the second data stream to synthesize a continuous scenario vector; the discrete fact data packets and the continuous scenario vectors are mutually exclusive in data structure.

[0104] The data interaction device provided in the embodiments of this disclosure firstly receives natural language data from a user in a receiving unit 301; secondly, a separation unit 302 performs feature separation on the natural language data to generate discrete fact data packets with non-overlapping main features and continuous scenario vectors; thirdly, an update unit 303 updates a pre-set fact database based on the discrete fact data packets to obtain an updated fact database; next, an attenuation unit 304 attenuates the continuous scenario vectors to obtain and writes the attenuated scenario vectors into a pre-set time-series database; then, a calculation unit 305 calculates dynamic view parameters based on the current scenario vectors in the time-series database; finally, a reconstruction unit 306 performs runtime reconstruction on the data in the updated fact database based on the dynamic view parameters to generate and output response data. This device breaks the chaotic state of natural language and no longer stops at the natural language processing level. Through hierarchical state compression, it transforms the irreversible information entropy increase process into an ordered state machine flow. It can rigidly lock facts like a traditional database, effectively preventing information drift, and can perceive scenarios like a human assistant, realizing dynamic empathy. This enables the system corresponding to the data interaction method to operate stably in complex environments for a long time.

[0105] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0106] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0107] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0108] like Figure 4 As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0109] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0110] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as data interaction methods. For example, in some embodiments, the data interaction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the data interaction method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the data interaction method by any other suitable means (e.g., by means of firmware).

[0111] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0112] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data interaction device, such that when executed by the processor or controller, the program code causes the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0113] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0114] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0115] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0116] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0117] The foregoing description of specific exemplary embodiments of this disclosure is for illustrative and explanatory purposes. These descriptions are not intended to limit this disclosure to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this disclosure and their practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this disclosure, as well as various different choices and variations. The scope of this disclosure is intended to be defined by the claims and their equivalents.

Claims

1. A data interaction method, the method comprising: Receive natural language data from users; The natural language data is subjected to feature separation to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; Based on the discrete fact data packets, the pre-set fact database is updated to obtain the updated fact database; The continuous scenario vector is written into a pre-set time series database, and the continuous scenario vector is attenuated based on the current system time and the timestamp of the historical scenario vector in the time series database to obtain the current scenario vector. Calculate the dynamic view parameters based on the current scenario vector; Based on the dynamic view parameters, the data in the updated fact database is reconstructed at runtime to generate and output response data.

2. The method according to claim 1, wherein, The calculation of dynamic view parameters based on the current scenario vector includes: The current scenario vector is processed by a weighted moving average to obtain the system weighted state value; Based on the weighted state values ​​of the system, dynamic view parameters are obtained through an anti-entropy adjustment strategy.

3. The method according to claim 2, wherein, The dynamic view parameters obtained based on the system's weighted state values ​​through an anti-entropy adjustment strategy include: Obtain preset opening threshold and closing threshold, wherein the opening threshold is greater than the closing threshold; Compare the system weighted state value with the on threshold and the off threshold respectively; In response to the system weighted state value being greater than the enable threshold, or in response to the system operating mode being dimensionality reduction protection mode in the previous moment and the system weighted state value being greater than the disable threshold, the system operating mode at the current moment is determined to be dimensionality reduction protection mode; In response to the system weighted state value being less than or equal to the shutdown threshold, the system operating mode at the current moment is determined to be full information mode; Based on the determined system operation mode, the corresponding set of control parameters is retrieved from the preset parameter mapping relationship and used as the dynamic view parameters; wherein, the set of control parameters corresponding to the dimensionality reduction protection mode represents the compressed information capacity, and the set of control parameters corresponding to the full information mode represents the full information capacity.

4. The method according to claim 1, wherein, The step of updating the pre-set fact database based on the discrete fact data packet to obtain the updated fact database includes: Extract the task description text from the discrete fact data packet and convert the task description text into a high-dimensional semantic vector; Calculate the cosine similarity between the high-dimensional semantic vector and the task vectors already stored in the fact database; In response to the cosine similarity being greater than a preset semantic overlap threshold, based on the constraints in the discrete fact data packet, the fields of the corresponding existing task records in the fact database are updated to obtain the updated fact database. In response to the cosine similarity being less than the semantic overlap threshold, the discrete fact data packet is written as a new record into the fact database to obtain an updated fact database.

5. The method according to any one of claims 1-4, wherein, The process of attenuating the continuous scenario vectors based on the timestamps of the current system time and the historical scenario vectors in the time-series database to obtain the current scenario vector includes: The time difference is calculated based on the current system time and the timestamp of the historical scenario vector in the time series database. Based on the time difference, the emotional valence feature value and arousal feature value in the continuous scenario vector are weighted using an exponential decay function to obtain decayed feature values. The decayed feature values ​​are combined into a current scenario vector, and the current scenario vector and the current system time are written into the time series database. The decay coefficient of the exponential decay function is inversely proportional to a preset forgetting time constant, so that the decayed scenario vector converges numerically to a preset neutral baseline state.

6. The method according to any one of claims 1-4, wherein, The step of reconstructing the data in the updated fact database at runtime based on the dynamic view parameters, and generating and outputting response data, includes: The dynamic view parameters are analyzed to obtain the information capacity limit, time horizon window value, and granularity level. Time filtering conditions are constructed based on the time horizon window value, and the corresponding task level field is selected based on the granularity level. The time filtering conditions and the task level fields are pushed down to the database query layer of the updated fact database; Construct a query statement that includes an entry number limit clause, wherein the parameters of the entry number limit clause are determined by the information capacity limit value; The query statement is executed, and a subset of data that matches the current context is extracted from the updated fact database as response data, which is then output.

7. The method according to claim 1, wherein, The step of performing feature separation on the natural language data to generate discrete fact data packets and continuous context vectors with non-overlapping main features includes: The natural language data is input into a pre-trained large language model; The decoding process of the large language model is constrained by a predefined text data pattern, and the first data stream and the second data stream are generated in parallel. Stop word filtering is performed on the first data stream to remove fields containing words with subjective emotions, thereby generating the discrete fact data package; The cognitive load index is calculated based on the textual physical features of the natural language data, and the continuous scenario vector is synthesized by combining the emotion classification probability in the second data stream; the discrete fact data package and the continuous scenario vector are not contained in each other in terms of data structure.

8. A data interaction device, the device comprising: The receiving unit is configured to receive the user's natural language data; The separation unit is configured to perform feature separation on the natural language data to generate discrete fact data packets and continuous scenario vectors with non-overlapping main features; The update unit is configured to update a pre-set fact database based on the discrete fact data packet to obtain an updated fact database. The attenuation unit is configured to write the continuous scenario vector into a pre-set time series database, and perform attenuation processing on the continuous scenario vector based on the current system time and the timestamp of the historical scenario vector in the time series database to obtain the current scenario vector. The calculation unit is configured to calculate dynamic view parameters based on the current scenario vector; The reconstruction unit is configured to perform runtime reconstruction of the data in the updated fact database based on the dynamic view parameters, and generate and output response data.

9. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images