A Smart Generation System and Method for WebXR Immersive Applications Based on a Large Language Model

Abstract

This invention relates to the field of intelligent generation technology, specifically to an intelligent generation system and method for WebXR immersive applications based on a large language model. The system includes: a semantic parsing module, a scene constraint module, an execution structure module, a stability determination module, and a code generation module. In this invention, after text or voice input enters the processing chain through the generation interface, parsing and sorting are completed at the sentence granularity. Spatial objects, interactive behaviors, and operation method information are identified simultaneously, reducing semantic ambiguity caused by natural expression and forming a clear order of interactive constraints. This gives the scene intent a traceable structural foundation. On this basis, triggering conditions, state maintenance, and resource requirements are added to construct an execution structure with temporal consistency. The execution requirements are determined in conjunction with the capabilities of the operating environment, so that the interactive logic and operating conditions are synchronously mapped to code expression, reducing the problem of manual adaptation and inconsistency with the environment, and improving the generation efficiency and operational stability of WebXR applications.

Description

Technical Field

[0001] This invention relates to the field of intelligent generation technology, and in particular to an intelligent generation system and method for WebXR immersive applications based on a large language model. Background Technology

[0002] The field of intelligent generation technology refers to a set of methods that use computer models to represent, reason about, and combine multi-source information to automatically form structured or semi-structured knowledge content. Its core aspects include knowledge acquisition, knowledge representation, knowledge association, and knowledge construction. This involves parsing textual data, monitoring data, charts, and standard documents, using rule-based constraints, semantic mapping relationships for extraction, and knowledge organization to uniformly model and collaboratively construct dispersed and heterogeneous information, thereby forming a domain knowledge system that can be processed by computers and continuously expanded. Specifically, the collaborative construction of knowledge in the traditional hydrogeological and environmental geology field refers to a method of manually organizing and classifying survey reports, exploration data, monitoring records, and technical specifications related to hydrogeological, engineering geological, and environmental geology technologies. This involves manually annotating stratigraphic structures, hydrological parameters, engineering characteristics, and environmental elements to summarize fields, completing the correspondence and summarization between different professional knowledge through a manually set classification system and association rules, and achieving the collaborative construction of knowledge in the hydrogeological and environmental geology field based on existing industry standards and expert experience.

[0003] Traditional knowledge collaboration in the field of hydrogeology and environmental geology relies on manual sorting and classification. Field summarization, classification system maintenance, and association rule setting are all driven by personnel experience. When faced with heterogeneous data from the same source, inconsistencies in terminology and boundary drift are prone to occur. In cross-personnel collaboration scenarios, multiple names for the same concept and multiple sets of terms for the same indicator are common, leading to increased costs for subsequent aggregation and alignment. Manual annotation is sensitive to data scale and update frequency. When monitoring records continue to grow or standard clauses are updated, repetitive sorting and retrospective correction consume a lot of time. Update delays create a time lag between the knowledge system and the latest data, thus affecting the accuracy of retrieval and reuse. Association rules are mainly statically set. When encountering new professional branches or new combinations of elements, rule expansion relies on centralized expert review. The expansion cycle is long and it is difficult to cover all marginal cases, which can easily lead to missed associations or misclassifications, causing knowledge network breaks or mislinks. In practical applications, when faced with the same survey report and monitoring forms, different personnel may have different granularities in extracting stratigraphic structure, hydrological parameters, and engineering characteristics. This can lead to missing or conflicting fields during the summarization process, further resulting in unstable model input, repeated revisions of statistical standards, and frequent rollbacks of deliverables. Consequently, the project delivery schedule is constrained by the manual processing throughput and consistency control capabilities. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a WebXR immersive application intelligent generation system and method based on a large language model. The technical solution is as follows: On the one hand, a WebXR immersive application intelligent generation system based on a large language model is provided. This system is implemented in a browser runtime environment that supports WebXR and is used to generate WebXR application webpage code that can be directly loaded and executed by the browser. The WebXR application webpage code includes HTML and JavaScript code. The system includes the following steps:

[0005] The semantic parsing module obtains the text or voice content entered by the user in the WebXR application's generation interface, parses the input information sentence by sentence, identifies spatial object information, action information, and operation information, and forms a set of semantic descriptions of the WebXR scene. The scene constraint module checks the semantic descriptions based on the WebXR scene semantic description set to determine whether they include both spatial object information and interactive behavior information. For semantic descriptions that meet the conditions, it establishes a correspondence between objects and interactions to form a WebXR scene interaction constraint set. The execution structure module generates execution node descriptions based on the WebXR scene interaction constraint set, supplements the execution nodes with interaction trigger sources, state maintenance conditions and resource usage requirements, and connects the execution nodes in the constraint order to construct a WebXR execution structure description sequence; The stability determination module calls the WebXR execution structure description sequence to obtain the running capability information of the browser environment and WebXR session, determines the resource requirements and interaction characteristics of the execution nodes, distinguishes between nodes that can be run directly and nodes that depend on running conditions, and generates an execution stability mark sequence. The code generation module maps execution nodes to WebXR interactive code based on the execution stability marker sequence and the WebXR execution structure description sequence. It generates directly executable code for nodes that can be run directly and generates code including execution judgment for nodes that depend on running conditions, and outputs WebXR application code entities.

[0006] As a further aspect of the present invention, the WebXR scene semantic description set includes spatial object lexics, interactive action lexics, and input operation lexics; the WebXR scene interaction constraint set includes object-action binding relationships, constraint order indexes, and binding confidence flags; the WebXR execution structure description sequence includes execution node descriptions, trigger source descriptions, state maintenance condition descriptions, resource usage requirement descriptions, and node connection topologies; the execution stability flag sequence includes capability satisfaction flags, condition dependency flags, and stability order indexes; and the WebXR application code entities include directly executable code segments, condition judgment code segments, and sequential combination scripts.

[0007] As a further aspect of the present invention, the semantic parsing module includes:

[0008] The input acquisition submodule acquires the text or voice content entered by the user in the WebXR application generation interface, reads the text character stream and speech transcription sequence sentence by sentence, detects the start and end positions of each sentence and records the sentence order index, collects the sentence length value and word count value, completes the order marking based on the sentence order index, and generates the sentence order mark quantity.

[0009] Based on the sentence order markers, the semantic recognition submodule performs word comparison and label judgment on each sentence content. It matches and counts the words associated with spatial objects with the preset object vocabulary, matches and counts the words associated with interactive actions with the action instruction table, and matches and counts the words associated with input methods with the operation identifier table. The three types of matching values ​​are obtained and merged to obtain the semantic classification count value.

[0010] The sequential integration submodule concatenates the spatial object description information, action description information, and operation description information in the original order of appearance according to the semantic classification count value and the corresponding sentence order index. It then assigns numbers to the description codes and forms a sequence mapping to generate a WebXR scene semantic description set.

[0011] As a further aspect of the present invention, the scene constraint module includes:

[0012] The semantic inspection submodule obtains the set of semantic descriptions of the WebXR scene, checks each semantic description one by one, reads the spatial object description information marker value and the interactive behavior description information marker value in the description, judges the two types of marker values ​​within the same semantic description, identifies the semantic item sequence number that meets the judgment condition based on the co-occurrence judgment benchmark value, and generates a co-occurrence judgment sequence number.

[0013] Based on the co-occurrence judgment sequence number, the relationship recording submodule calls the corresponding semantic description content, extracts the spatial object identifier value and the interaction behavior identifier value, pairs the two types of identifier values ​​and records the mapping number, establishes an object and interaction pairing record for each identified semantic description, and obtains the object interaction corresponding quantity;

[0014] The sequential summary submodule arranges the corresponding object interaction quantities according to the original order of appearance of the semantic description, arranges the paired records in sequence to form a continuous sequence, and summarizes the sequence by number to generate a WebXR scene interaction constraint set.

[0015] As a further aspect of the present invention, in the WebXR scene semantic description set obtained by the semantic inspection submodule, each semantic description contains at least one spatial object description information marker value and at least one interactive behavior description information marker value; the co-occurrence judgment benchmark value is limited to a threshold value for determining the number of spatial object description information marker values ​​and interactive behavior description information marker values ​​that appear simultaneously within the same semantic description, wherein the number threshold value is not less than one; the process of identifying the semantic entry sequence number that meets the judgment condition based on the co-occurrence judgment benchmark value is specifically limited to recording the original sequence number of the corresponding semantic description in the WebXR scene semantic description set only when the number of spatial object description information marker values ​​and the number of interactive behavior description information marker values ​​detected within the same semantic description both meet the number threshold value; the process of pairing the two types of marker values ​​and recording the mapping number is specifically limited to generating a one-to-one pairing relationship in the order of spatial object marker value first and interactive behavior marker value last, and assigning a unique incremental mapping number to each pairing relationship; the process of arranging according to the original order of appearance of the semantic description is specifically limited to sorting according to the original sequence number from small to large, and sequentially connecting the sorted pairing records to form a continuous number sequence.

[0016] As a further aspect of the present invention, the execution structure module includes:

[0017] The node generation submodule obtains the WebXR scene interaction constraint set, reads the corresponding records of object identifier value and interaction identifier value one by one, generates an execution node number for each set of corresponding records, records the combination relationship between object number and interaction number, detects the continuity of number and marks the node position index, completes the node entry registration according to the node position index, and generates the execution node number quantity.

[0018] Based on the execution node number, the condition supplementation submodule calls the corresponding object number and interaction number, collects the interaction trigger source identifier value, the state maintenance condition identifier value and the resource usage identifier value, merges the three types of identifier values ​​at the node level and records the attribute sequence number, and forms an attribute entry set for each node to obtain the node attribute combination quantity;

[0019] The sequential connection submodule connects the execution nodes one after the other according to the original order of the interaction constraint set based on the combination of node attributes and the node position index. It records the connection number and sequence mapping value between nodes, forms a chain sequence of consecutive nodes and summarizes the numbers, and establishes a WebXR execution structure description sequence.

[0020] As a further aspect of the present invention, the stability determination module includes:

[0021] The capability acquisition submodule calls the WebXR execution structure description sequence to obtain the runtime capability information provided by the browser environment and the runtime capability information provided by the WebXR session. It collects the number of computing units, the maximum memory usage, the graphics support level, and the session permission identifier. It standardizes the format and registers the capability values, establishes a capability item index table, and generates a quantity of runtime capability parameters.

[0022] Based on the aforementioned operational capability parameters, the node determination submodule reads the resource usage requirements and interaction characteristic requirements for each execution node, compares the node resource requirement value with the corresponding capability upper limit value, marks records whose node requirement value does not exceed the capability upper limit value as directly runnable, and marks records whose node requirement value exceeds the capability upper limit value as conditionally dependent, and summarizes the node determination identifiers to obtain the node stability identifier quantity.

[0023] The sequence arrangement submodule arranges the execution node determination identifiers according to the original connection order based on the node stability identifier quantity and the execution structure description sequence node sequence number, maps the stability identifiers and node numbers sequentially and records them to generate an execution stability mark sequence.

[0024] As a further aspect of the present invention, the number of computing units is limited to the minimum number of logical processing units that can be allocated simultaneously in the browser environment and the WebXR session; the upper limit of memory usage is limited to the minimum available memory capacity identified in the runtime capability information provided by the browser environment and the runtime capability information provided by the WebXR session; the process of unifying the format and registering the capability values ​​is specifically limited to converting the number of computing units, the upper limit of memory usage, the graphics support level, and the session permission identifier into a unified data type and writing them into the capability entry index table according to a preset field order; the process of comparing the node resource requirement value with the corresponding capability upper limit value is specifically limited to comparing the resource usage requirement value with the number of computing units, the upper limit of memory usage, and the graphics support level value item by item, and recording the condition dependency flag when any comparison result exceeds the corresponding capability upper limit value; the process of sequentially mapping and combining the stability identifier with the node number is specifically limited to using the sequential number of the execution structure description sequence node as the primary key, and writing the corresponding node stability identifier into the execution stability flag sequence in a one-to-one correspondence.

[0025] As a further aspect of the present invention, the code generation module includes:

[0026] The node mapping submodule reads the execution node number and stability identifier value one by one according to the execution stability marker sequence and the WebXR execution structure description sequence. For each execution node, it calls the object identifier, interaction identifier and node attribute entry, converts the node attribute entry into the corresponding interaction instruction parameter set, records the instruction sequence number and parameter length value, and generates the node instruction mapping quantity.

[0027] Based on the node instruction mapping quantity and execution stability flag value, the condition generation submodule determines whether the node identifier is directly runnable or conditionally dependent. For directly runnable nodes, it extracts the instruction parameter set to form a code fragment number. For conditionally dependent nodes, it overlays the running condition judgment flag and generates a conditional code fragment number. The two types of code fragments are numbered and registered to obtain the code fragment identifier quantity.

[0028] The sequential combination submodule concatenates the code fragments in sequence according to the code fragment identifier and the sequential number of the execution structure description sequence node, records the fragment connection order value and the total length count value, summarizes the overall combination result with a number, and outputs the WebXR application code entity.

[0029] On the other hand, the intelligent generation system and method for WebXR immersive applications based on a large language model, wherein the intelligent generation system and method for WebXR immersive applications based on a large language model is executed based on the aforementioned intelligent generation system for WebXR immersive applications based on a large language model, includes the following steps: S1: Obtain the text or voice content entered by the user in the WebXR application generation interface, read the input content sentence by sentence, identify the spatial object description information used to describe the WebXR scene, the action description information used to describe the XR interaction behavior, and the operation description information used to describe the input method, and organize the description information according to the original order of appearance to form a WebXR scene semantic description set. S2: Based on the WebXR scene semantic description set, check each semantic description to determine whether spatial object description information and interactive behavior description information appear simultaneously in the semantic description. For semantic descriptions that meet the judgment conditions, record the correspondence between the object and the interaction, and summarize the correspondence in the order of description to form a WebXR scene interaction constraint set. S3: Based on the WebXR scene interaction constraint set, generate an execution node description for each group of objects and interaction correspondences, supplement the execution node with the interaction trigger source, state maintenance conditions and resource usage requirements, and connect the execution nodes in the order of the interaction constraint set to establish a WebXR execution structure description sequence; S4: Call the WebXR execution structure description sequence to obtain the running capability information provided by the browser environment and WebXR session, judge the resource usage requirements and interaction characteristic requirements described by each execution node, distinguish between execution nodes that meet the running capability conditions and execution nodes that depend on the target running conditions, and organize the judgment results according to the execution structure order to generate an execution stability mark sequence. S5: Based on the execution stability marker sequence and the WebXR execution structure description sequence, the execution nodes are mapped to the corresponding WebXR interactive code content. For execution nodes that meet the running capability conditions, code segments that can be executed directly are generated. For execution nodes that depend on the target running conditions, code segments with running condition judgments are generated. The code segments are combined in the execution order to output the WebXR application code entity.

[0030] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, after text or voice input enters the processing chain through the generation interface, it is parsed and sorted at the statement granularity, and spatial objects, interactive behaviors, and operation methods are identified simultaneously. This reduces semantic ambiguity caused by natural expression, forms a clear order of interactive constraints, and gives the scene intent a traceable structural foundation. On this basis, triggering conditions, state maintenance, and resource requirements are added to build an execution structure with temporal consistency. The execution requirements are judged in combination with the capabilities of the runtime environment, so that the interactive logic and runtime conditions are synchronously mapped to code expression, reducing the problems of manual adaptation and inconsistency with the environment, and improving the generation efficiency and runtime stability of WebXR applications. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the intelligent generation system for WebXR immersive applications based on a large language model provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention; Figure 3 This is a flowchart of the semantic parsing module in this invention; Figure 4 This is a flowchart of the scene constraint module in this invention; Figure 5 This is a flowchart of the execution structure module in this invention; Figure 6This is a flowchart of the stability determination module in this invention; Figure 7 This is a flowchart of the code generation module in this invention; Figure 8 This is a flowchart of the intelligent generation system and method for WebXR immersive applications based on a large language model provided in the embodiments of the present invention. Detailed Implementation

[0033] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0034] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0035] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0036] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0037] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0038] This invention provides an intelligent generation system for WebXR immersive applications based on a large language model, such as... Figure 1-2 The diagram shown illustrates an intelligent generation system for WebXR immersive applications based on a large language model. The system includes:

[0039] The semantic parsing module acquires the text or voice content entered by the user in the WebXR application generation interface, reads the input content sentence by sentence, identifies spatial object description information used to describe the composition of the WebXR scene, action description information used to describe XR interaction behavior, and operation description information used to describe the input method, and organizes the description information according to the original order of appearance to form a set of semantic descriptions of the WebXR scene.

[0040] The scene constraint module is based on the WebXR scene semantic description set. It checks each semantic description to determine whether spatial object description information and interactive behavior description information appear simultaneously in the semantic description. For semantic descriptions that meet the judgment conditions, it records the correspondence between objects and interactions, and summarizes the correspondence in the order of description to form the WebXR scene interaction constraint set.

[0041] The execution structure module generates an execution node description for each group of objects and interactions based on the WebXR scene interaction constraint set. It supplements the execution node with the interaction trigger source, state maintenance conditions and resource usage requirements, and connects the execution nodes in the order of the interaction constraint set to establish a WebXR execution structure description sequence.

[0042] The stability determination module calls the WebXR execution structure description sequence to obtain the runtime capability information provided by the browser environment and WebXR session. It judges the resource usage requirements and interaction characteristic requirements described by each execution node, distinguishes between execution nodes that meet the runtime capability conditions and execution nodes that depend on the target runtime conditions, and organizes the judgment results according to the execution structure order to generate an execution stability mark sequence.

[0043] The code generation module maps execution nodes to corresponding WebXR interactive code content based on the execution stability marker sequence and the WebXR execution structure description sequence. For execution nodes that meet the running capability conditions, it generates directly executable code segments. For execution nodes that depend on the target running conditions, it generates code segments with running condition judgments. The code segments are combined in the execution order to output the WebXR application code entity.

[0044] The WebXR scene semantic description set includes spatial object lexics, interactive action lexics, and input operation lexics; the WebXR scene interaction constraint set includes object-action binding relationships, constraint order indexes, and binding confidence flags; the WebXR execution structure description sequence includes execution node descriptions, trigger source descriptions, state preservation condition descriptions, resource usage requirement descriptions, and node connection topology; the execution stability flag sequence includes capability satisfaction flags, condition dependency flags, and stability order indexes; and the WebXR application code entities include directly executable code segments, conditional judgment code segments, and sequential combination scripts.

[0045] Specifically, such as Figure 2 , 3 As shown, the semantic parsing module includes:

[0046] The input acquisition submodule acquires the text or voice content entered by the user in the WebXR application generation interface, reads the text character stream and speech transcription sequence sentence by sentence, detects the start and end positions of each sentence and records the sentence order index, collects the sentence length value and word count value, completes the order marking based on the sentence order index, and generates the sentence order mark quantity.

[0047] During the initialization phase, both text and voice input channels are monitored simultaneously. The text input channel directly receives character stream data from the input box of the generation interface, while the voice input channel receives voice stream from the browser's audio interface and outputs a speech-to-text character sequence by the built-in transcription engine. First, the two types of input are uniformly sorted and registered based on the character stream timestamp and the transcription sequence timestamp. Then, sentence-by-sentence segmentation is performed according to the terminator and silence interval. The terminator detection is limited to three types of punctuation: period, question mark, and exclamation mark. The silence interval is fixed at 400 milliseconds; when the continuous silence duration reaches this value, it is registered as a sentence boundary. After segmentation, a sentence order index is registered for each input sentence. The sentence order index starts from 1 and increments sequentially, and the start and end positions of the sentence in the original character stream are recorded simultaneously. The process involves several steps: First, character cleaning is performed on the clause content. The cleaning rules are limited to removing duplicate spaces, tabs, and semantically meaningless symbols, and converting full-width characters to half-width characters. After cleaning, term segmentation is performed on each sentence. The segmentation is based on a pre-set basic vocabulary. Content not matched by the vocabulary is split into segments with a minimum consecutive character length of 2. After obtaining the term sequence, the sentence length and term count for each sentence are recorded. The sentence length is recorded based on the number of characters after cleaning, and the term count is recorded based on the number of terms after segmentation. For example, in one actual data collection, four sentences were collected, with sentence indices from 1 to 4, corresponding to sentence lengths of 28, 31, 22, and 26, and term counts of 12, 14, 10, and 11, respectively. All of the above data are written into the collection record set with the sentence index as the primary key, forming a sentence sequence marker.

[0048] The semantic recognition submodule, based on sentence order markers, performs word comparison and label judgment on sentence-by-sentence content. It matches and counts the words associated with spatial objects with a pre-set object vocabulary, matches and counts the words associated with interactive actions with an action instruction table, and matches and counts the words associated with input methods with an operation identifier table. The three types of matching values ​​are obtained and merged to obtain the semantic classification count value.

[0049] First, the corresponding word sequence is read line by line according to the sentence order index. Within the same processing cycle, the spatial object lexicon, action instruction table, and operation identifier table are called to perform word matching and registration. During spatial object matching, each word in the word sequence is compared with the standardized words in the object lexicon for consistency. Simultaneously, the synonym mapping table is called to normalize the word mapping before comparison. Successful matches accumulate the spatial object matching count. During action instruction matching, the same logic is used to compare words with the action instruction table and accumulate the action matching count. During operation identifier matching, each word is compared with the operation identifier table and the operation matching count is accumulated. These three types of matching... During the matching process, if the same term appears multiple times in the same sentence, the number of occurrences is accumulated without deduplication. After completing the three-class matching, the spatial object matching count, action matching count, and operation matching count are written into the same semantic classification record, and the three types of values ​​are merged and registered. The merging logic is to add the three types of matching counts in sequence to form the semantic classification count. For example, in the sentence with sentence order index 2, the spatial object matching count is 3, the action matching count is 2, and the operation matching count is 1. The semantic classification count of the sentence is registered as 6, and the counts of each item and the merging result are written into the semantic recognition record set. This set still uses the sentence order index as the unique index identifier.

[0050] The sequential integration submodule concatenates spatial object description information, action description information, and operation description information in the original order of appearance according to the semantic classification count value and the corresponding sentence order index. It then encodes and numbers the descriptions and forms a sequence mapping to generate a set of WebXR scene semantic descriptions.

[0051] After receiving the semantic recognition record set, the system reads the corresponding records one by one in ascending order of sentence index. Its operational logic revolves around the structured assembly of semantic content. First, a description code number is generated for each semantic record. This number is generated using a fixed prefix followed by an incrementing number, with each incrementing number corresponding to a sentence index. Then, the system reads the spatial object matching term name, action instruction matching term name, and operation identifier matching term name corresponding to the sentence, and performs sequential assembly based on the order in which the original terms appear in the sentence. During assembly, fields are not padded across sentences. When the matching count for a certain category is 0, the corresponding field is written... An empty tag does not affect the writing of other fields; at the same time, the sentence start position number and end position number are written into the description header information to maintain the consistency with the original input location; after completing the single sentence concatenation, the submodule registers the description code number, sentence order index, three types of description fields and position information as a semantic description record; for example, in the aforementioned 4 sentences, 4 semantic description records are generated, with their description code numbers increasing sequentially, and the field order in each record is consistent with the word order in the original sentence; after all sentences are processed, the semantic description records are written into the webpage extended reality scene semantic description set in the order of description code numbers.

[0052] Specifically, such as Figure 2 , 4 As shown, the scene constraint module includes:

[0053] The semantic inspection submodule obtains a set of WebXR scene semantic descriptions, examines each semantic description one by one, reads the spatial object description information marker value and the interactive behavior description information marker value in the description, makes a judgment on the two types of marker values ​​within the same semantic description, identifies the semantic item sequence number that meets the judgment condition based on the co-occurrence judgment benchmark value, and generates a co-occurrence judgment sequence number.

[0054] Configured within the scene constraint module, its processing object is the set of semantic descriptions of web-extended real-world scenes. It reads semantic description records one by one according to their description code numbers, and simultaneously reads the spatial object description information marker value and the interaction behavior description information marker value within each record. During the judgment phase, an internal consistency judgment is performed using a co-occurrence judgment benchmark value. The benchmark value is fixed at 2, meaning that in the same semantic description record, both the spatial object matching count value and the action matching count value reach 2 or higher, and the corresponding description field is not empty. This benchmark value is solidified through offline experimental data, which comes from the manually labeled results of 1200 real generated instructions. When the benchmark value is 1, there are 96 misjudged entries; when the benchmark value is 2, there are 28 misjudged entries; and when the benchmark value is 3, there are 74 missed entries. The fixed benchmark value is 2. During online operation, semantic description records that meet the above conditions are marked as valid co-occurrence records. The sentence order index corresponding to their description code number is written into the co-occurrence judgment sequence number, and a judgment timestamp is recorded. For example, in the aforementioned four semantic description records, the records with sentence order indices 2 and 4 meet the co-occurrence condition.

[0055] The relationship recording submodule, based on the co-occurrence judgment sequence number, calls the corresponding semantic description content, extracts the spatial object identifier value and the interaction behavior identifier value, pairs the two types of identifier values ​​and records the mapping number, establishes an object and interaction pairing record for each identified semantic description, and obtains the object interaction corresponding quantity;

[0056] After receiving the co-occurrence judgment sequence number, the corresponding semantic description record is called in sequence, and the spatial object identifier value and interaction behavior identifier value are read from it. The spatial object identifier value is generated by the object specification term using a fixed hash rule, and the interaction behavior identifier value is generated by the action instruction specification term using the same rule. The hash rule uses a fixed salt value to ensure consistency across sessions. The two types of identifier values ​​are paired and registered, and a mapping number is generated for each pair. The mapping number is generated incrementally according to the processing order. When the same spatial object identifier value corresponds to different interaction behaviors in different semantic description records, the system will handle the situation. When the identifier value is not specified, independent mapping numbers are generated for each, and no merging process is performed. For example, in the record with sentence order index 2, the spatial object identifier value is 10021 and the interaction behavior identifier value is 20007. In the record with sentence order index 4, the spatial object identifier value is still 10021, but the interaction behavior identifier value is 20009. Mapping numbers 1 and 2 are generated respectively, and the corresponding sentence order indexes are retained in the records. All pairing results are registered as a set of object interaction correspondence quantities. Each record in the set contains a mapping number, a spatial object identifier value, an interaction behavior identifier value, and a source sentence order index.

[0057] The sequential summary submodule arranges the corresponding objects based on their associated sequence number information according to the original order of their semantic descriptions, concatenates the paired records in sequence to form a continuous sequence, numbers and summarizes the sequence, and generates a set of WebXR scene interaction constraints.

[0058] Taking the set of object interaction correspondences as the processing object, internally, it first performs a stable sorting of all paired records based on the source sentence order index. The sorting rule is to arrange them in ascending order of sentence order index, and records with the same sentence order index maintain their original generation order. Then, it performs continuous numbering rewriting on the sorted records, replacing the original mapping number with a continuous sequence number, which is generated from 1 to eliminate the index discontinuity problem caused by missing numbers in the middle. Then, it concatenates the paired records after continuous numbering in order to form an interaction constraint sequence, and generates an interaction constraint set number for this sequence. The set number is composed of a fixed prefix and the millisecond timestamp of the generation time. In the example above, the interaction constraint sequence of 2 records has continuous sequence numbers of 1 and 2, which is consistent with the sentence order indexes 2 and 4. The output is the webpage extended real-world scene interaction constraint set.

[0059] Specifically, such as Figure 2 , 5 As shown, the execution structure module includes:

[0060] The node generation submodule obtains the WebXR scene interaction constraint set, reads the corresponding records of object identifier value and interaction identifier value one by one, generates execution node number for each set of corresponding records, records the combination relationship between object number and interaction number, detects the continuity of number and marks the node position index, completes the node entry registration according to the node position index, and generates the execution node number quantity.

[0061] The system reads each consecutive record in the interaction constraint set and generates an execution node number for each record. The execution node number uses a two-segment structure: the first segment is taken from the truncated value of the interaction constraint set number, with a truncated length of 8 digits; the second segment is the node sequence number, which is consistent with the consecutive sequence number. Simultaneously, the system records the combination relationship between the object number and the interaction number, and writes the node position index into the node record, directly using the node sequence number. Then, it performs a node continuity check, which is performed by traversing the node position index incrementally. When a subsequent node position index is found to be not equal to the previous node position index plus 1, a breakpoint is registered. In the previous example, the interaction constraint set contains two consecutive records, generating node numbers 1 and 2, and no breakpoint is registered. All node records are aggregated into an execution node number set, where each record contains an execution node number, an object number, an interaction number, and a node position index.

[0062] The condition supplementation submodule, based on the execution node number, calls the corresponding object number and interaction number, collects the interaction trigger source identifier value, state maintenance condition identifier value and resource usage identifier value, merges the three types of identifier values ​​at the node level and records the attribute sequence number, and forms an attribute entry set for each node to obtain the node attribute combination quantity;

[0063] After reading the set of execution node numbers, the node records are called one by one according to the node position index, and the node attribute-related identifier values ​​are collected synchronously from the runtime environment listener and resource manager; the interaction trigger source identifier value is provided by the input event listener, which registers controller button, gesture event, and gaze ray hit as source numbers 1, 2, and 3 respectively; the state maintenance condition identifier value is provided by the state recorder, which counts the time span between the start and end of the interaction and maps it to condition numbers according to the interval: 0 to 500 milliseconds is mapped to 1, 501 to 2000 milliseconds is mapped to 2, 2001 to 8000 milliseconds is mapped to 3, 4, 5, 6, 7, 8, 9, 1 ... Mapping is 3, and mapping is 4 for values ​​greater than 8000 milliseconds; the resource usage identifier value is provided by the resource manager, which records the peak usage of model resources, texture resources and audio resources and counts them in megabytes; the three types of identifier values ​​are merged at the node level, and an attribute sequence number is generated for each node, which is consistent with the node position index; for example, in the node with node position index 1, the trigger source is a gesture event corresponding to source number 2, the state holding time span is 3200 milliseconds corresponding to condition number 3, and the peak resource usage is 240 megabytes for the model, 180 megabytes for the texture and 60 megabytes for the audio.

[0064] The sequential connection submodule connects the execution nodes one after the other according to the original order of the interaction constraint set based on the combination of node attributes and the node position index. It records the connection number and sequence mapping value between nodes, forms a chain sequence of consecutive nodes and summarizes the numbers, and establishes a WebXR execution structure description sequence.

[0065] The system constructs connections between nodes based on node attribute records and node position indices. Internally, it reads node records in ascending order of node position indices and generates connection numbers for adjacent nodes. The connection number is registered by combining the previous node number and the next node number. During the connection process, it records sequence mapping values, which are used to identify the connection relationship between preceding and following nodes. When a breakpoint is registered during the node generation stage, a placeholder node is inserted at the breakpoint position and the connection relationship is split, but this logic is not triggered. After the connection is completed, all node connection relationships are written into a chain structure in sequence, and a chain number is generated for the chain structure, which is generated incrementing from 1. The output webpage extends to display the execution structure description sequence, and each record in the sequence contains the execution node number, node attribute record, and connection number.

[0066] Specifically, such as Figure 2 , 6As shown, the stability determination module includes:

[0067] The capability acquisition submodule calls the WebXR execution structure description sequence to obtain the runtime capability information provided by the browser environment and the WebXR session. It collects the number of computing units, the maximum memory usage, the graphics support level, and the session permission identifier. It standardizes the format and registers the capability values, establishes a capability item index table, and generates a quantity of runtime capability parameters.

[0068] The system obtains runtime capability-related parameters through the browser environment interface and session interface. The number of computing units is returned by the environment interface based on the number of logical cores. The maximum memory usage is determined by the browser's single-tab allocable limit and the device's available memory limit, taking the smaller value. The graphics support level is generated by combining the maximum texture size and the maximum number of texture units returned by the graphics interface. The session permission identifier is returned by the session interface after permission confirmation. The above capability parameters are processed in a standardized format and registered with capability numbers. For example, in an actual data acquisition, the number of computing units is 8, the maximum memory usage is 2048 megabytes, and the maximum texture size is 8192. The corresponding graphics support level is registered as 2, and the session permission identifier is registered as allowed. All capability parameters are written into the capability entry index set.

[0069] The node determination submodule reads the resource usage requirements and interaction characteristic requirements for each execution node based on the number of runtime capability parameters. It compares the node resource requirement value with the corresponding capability limit value and marks the records whose node requirement value does not exceed the capability limit value as directly runnable, and marks the records whose node requirement value exceeds the capability limit value as conditional dependency. The node determination identifiers are then summarized to obtain the number of node stability identifiers.

[0070] After receiving the set of capability entry indexes, the node attribute records in the execution structure description sequence are read one by one, and the node resource requirements and runtime capability parameters are compared and judged. The node resource requirement value is obtained by adding the peak usage of various resources in the node attribute record and comparing it with the upper limit of memory usage. When the resource requirement value does not exceed the upper limit of memory usage and the texture size requirement does not exceed the maximum texture size corresponding to the graphics support level, it is registered as directly runnable. When either condition is not met, it is registered as a conditional dependency. For example, in the node with node position index 1, the total resource requirement is 480 megabytes, which is less than 2048 megabytes, and the texture size requirement is 4096, which is less than 8192, so it is registered as directly runnable. In the node with node position index 2, the total resource requirement is 2310 megabytes, which is greater than 2048 megabytes, so it is registered as a conditional dependency and the excess item is recorded as memory usage.

[0071] The sequence arrangement submodule arranges the execution node judgment identifiers according to the original connection order based on the node stability identifier quantity and the execution structure description sequence node sequence number, maps the stability identifiers and node numbers sequentially and records them to generate an execution stability mark sequence.

[0072] After receiving the node determination result, the execution order of all node records is processed according to the established execution structure description sequence. First, using the node position index as the unique order benchmark, the execution node number, node stability identifier value, and corresponding over-limit item record are retrieved one by one, and the above information is written into the sequence mapping cache. During the cache establishment process, no merging or pruning of node content is performed; only the arrangement order of node records is linearly arranged to ensure consistency with existing node connection relationships. When the node stability identifier value is directly runnable, a stability flag is written into the corresponding mapping record. When the node stability identifier value is conditional dependency, an over-limit item array and an over-limit type number are written simultaneously. The over-limit type number is used for differentiation. Memory usage exceeds limits, graphics capabilities are limited, or session permissions are restricted. After completing the sequential mapping of all node records, the records in the mapping cache are renumbered to form an execution stability marker sequence. The numbers start from 1 and increment sequentially, maintaining a one-to-one correspondence with the node order. For example, in the aforementioned node determination results, the record with node position index 1 is marked as directly executable, and the record with node position index 2 is marked as conditionally dependent with memory usage exceeding limits. These markers are continuously registered in the execution stability marker sequence according to the original connection order. The generated execution stability marker sequence serves as the stability basis for subsequent instruction mapping and condition generation. Its internal record structure maintains node-level independence and does not involve cross-node state transfer.

[0073] Specifically, such as Figure 2 , 7 As shown, the code generation module includes:

[0074] The node mapping submodule reads the execution node number and stability identifier value one by one according to the execution stability marker sequence and the WebXR execution structure description sequence. For each execution node, it calls the object identifier, interaction identifier and node attribute entry, converts the node attribute entry into the corresponding interaction instruction parameter set, records the instruction sequence number and parameter length value, and generates the node instruction mapping quantity.

[0075] An index mapping relationship between node numbers and node attribute records is established based on the execution stability marker sequence and the WebXR execution structure description sequence. This mapping relationship uses the execution node number as the primary key to quickly locate object identifier values, interaction identifier values, and node attribute entries during the generation process. During the mapping execution phase, the execution node numbers are read sequentially according to the execution stability marker sequence, and the corresponding node attribute records are retrieved through the index mapping relationship. The object identifier value is converted into resource positioning parameters, which include model resource path identifiers, material resource path identifiers, and an initial pose parameter set. The initial pose parameter set consists of position parameters and orientation parameters, with position parameters in meters and orientation parameters stored in quaternion form. Simultaneously, the interaction identifier value is mapped to an interaction command template. The instruction template identifies the corresponding preset instruction parameter structure, which includes an event listening parameter segment, a state update parameter segment, and a resource loading parameter segment. During the generation of the event listening parameter segment, corresponding listening parameters are generated based on the trigger source identifier value. When the trigger source identifier value is 1, controller button listening parameters are generated; when it is 2, gesture recognition listening parameters are generated; and when it is 3, gaze ray hit listening parameters are generated. In the state update parameter segment, a duration parameter is generated based on the state maintenance condition identifier value. When the condition identifier value is 3, the duration parameter is registered as 5000 milliseconds. In the resource loading parameter segment, the peak resource usage record in the node attribute record is read and a loading budget parameter is generated. The above parameters are summarized to form a node instruction parameter set, and the instruction sequence number and parameter length value are registered.

[0076] The condition generation submodule determines whether a node is directly runnable or conditionally dependent based on the node instruction mapping quantity and execution stability flag value. For directly runnable nodes, it extracts the instruction parameter set to form a code fragment number. For conditionally dependent nodes, it overlays the running condition judgment flag and generates a conditional code fragment number. The two types of code fragments are numbered and registered to obtain the code fragment identification quantity.

[0077] After completing the registration of the node instruction parameter set, the execution conditions of each node are generated and processed according to the execution stability flag sequence. When the corresponding stability flag value is directly runnable, the node instruction parameter set is directly registered as the code fragment parameter source, and a code fragment number consistent with the node order is generated. When the stability flag value is conditional dependency, the condition processing logic is entered. First, the over-limit item record is read and the over-limit type is identified. When the over-limit type is memory usage, the difference calculation is performed between the total peak resource usage value in the node attribute record and the upper limit of memory usage registered in the capacity acquisition stage, resulting in an over-limit of 262 megabytes. Subsequently, the deloadable resource usage record is read and budget deduction processing is performed in order of resource type. The deduction order is fixed as texture resources first, followed by audio resources. When the original budget for texture resources is 600 megabytes, it is reduced by 262 megabytes and updated to 338 megabytes. After the budget adjustment is completed, the texture size parameter in the node instruction parameter set is updated synchronously, changing the original 8192 to 4096. When the over-limit type is limited graphics capabilities, texture size downgrade mapping is directly performed. When the over-limit type is limited session permissions, the immersive session parameter is adjusted to an inline session parameter. After completing the corresponding condition processing, the updated instruction parameter set is registered as a condition code snippet, and a new code snippet number is generated. All generated code snippet numbers are uniformly summarized into a code snippet identifier set.

[0078] The sequential assembly submodule concatenates code fragments in order according to the code fragment identifier and the sequential number of the execution structure description sequence node, records the fragment connection order value and the total length count value, summarizes the overall number of the assembly result, and outputs the WebXR application code entity.

[0079] After obtaining the complete set of code fragment identifiers, the combination process is performed according to the node order mapping records in the execution structure description sequence. During the combination initialization phase, a common dependency segment parameter set is first generated, containing resource index initialization parameters, event bus initialization parameters, and session entry initialization parameters, and written to the combination start position. During combination execution, code fragment numbers are read sequentially according to node order, and the corresponding code fragment parameter sets are appended to the combination buffer in turn. Strict consistency of node order is maintained during appending, and no rearrangement or merging is performed. After each append operation, the cumulative character length value in the combination buffer is counted in real time, based on the number of characters in the generated code and recorded in bytes. When a conditional code fragment is encountered, the condition judgment identifier and the main branch code are written to the buffer together, maintaining the integrity of the condition judgment structure. After all code fragments are appended, an integrity check is performed on the combination buffer, limited to code fragment number continuity and parameter missing detection. After the check passes, an application code entity number is generated for the combination result, and the code entity number, cumulative character length value, and node count are recorded in the output registration set, completing the generation and output of the webpage extended real-world application code entity.

[0080] Please see Figure 8 The WebXR immersive application intelligent generation system and method based on a large language model is executed based on the aforementioned WebXR immersive application intelligent generation system based on a large language model, and includes the following steps: S1: Obtain the text or voice content entered by the user in the WebXR application generation interface, read the input content sentence by sentence, identify the spatial object description information used to describe the WebXR scene, the action description information used to describe the XR interaction behavior, and the operation description information used to describe the input method, and organize the description information according to the original order of appearance to form a WebXR scene semantic description set. S2: Based on the WebXR scene semantic description set, check each semantic description to determine whether spatial object description information and interactive behavior description information appear simultaneously in the semantic description. For semantic descriptions that meet the judgment conditions, record the correspondence between objects and interactions, and summarize the correspondence in the order of description to form a WebXR scene interaction constraint set. S3: Based on the WebXR scene interaction constraint set, generate an execution node description for each group of objects and interaction correspondences, supplement the execution node with the interaction trigger source, state maintenance conditions and resource usage requirements, and connect the execution nodes in the order of the interaction constraint set to establish a WebXR execution structure description sequence; S4: Call the WebXR execution structure description sequence, obtain the runtime capability information provided by the browser environment and WebXR session, judge the resource usage requirements and interaction characteristic requirements described by each execution node, distinguish between execution nodes that meet the runtime capability conditions and execution nodes that depend on the target runtime conditions, and organize the judgment results according to the execution structure order to generate an execution stability mark sequence. S5: Based on the execution stability marker sequence and the WebXR execution structure description sequence, the execution nodes are mapped to the corresponding WebXR interactive code content. For execution nodes that meet the running capability conditions, code segments that can be executed directly are generated. For execution nodes that depend on the target running conditions, code segments with running condition judgments are generated. The code segments are combined in the execution order to output the WebXR application code entity.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A WebXR immersive application intelligent generation system based on a large language model, characterized in that, The system is implemented in a browser runtime environment that supports WebXR and is used to generate WebXR application webpage code that can be directly loaded and executed by the browser. The WebXR application webpage code includes HTML and JavaScript code, and the system includes the following steps: The semantic parsing module obtains the text or voice content entered by the user in the WebXR application's generation interface, parses the input information sentence by sentence, identifies spatial object information, action information, and operation information, and forms a set of semantic descriptions of the WebXR scene. The scene constraint module checks the semantic descriptions based on the WebXR scene semantic description set to determine whether they include both spatial object information and interactive behavior information. For semantic descriptions that meet the conditions, it establishes a correspondence between objects and interactions to form a WebXR scene interaction constraint set. The execution structure module generates execution node descriptions based on the WebXR scene interaction constraint set, supplements the execution nodes with interaction trigger sources, state maintenance conditions and resource usage requirements, and connects the execution nodes in the constraint order to construct a WebXR execution structure description sequence; The stability determination module calls the WebXR execution structure description sequence to obtain the running capability information of the browser environment and WebXR session, determines the resource requirements and interaction characteristics of the execution nodes, distinguishes between nodes that can be run directly and nodes that depend on running conditions, and generates an execution stability mark sequence. The code generation module maps execution nodes to WebXR interactive code based on the execution stability marker sequence and the WebXR execution structure description sequence. It generates directly executable code for nodes that can be run directly and generates code including execution judgment for nodes that depend on running conditions, and outputs WebXR application code entities.

2. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 1, characterized in that: The WebXR scene semantic description set includes spatial object lexics, interactive action lexics, and input operation lexics; the WebXR scene interaction constraint set includes object-action binding relationships, constraint order indexes, and binding confidence flags; the WebXR execution structure description sequence includes execution node descriptions, trigger source descriptions, state preservation condition descriptions, resource usage requirement descriptions, and node connection topology; the execution stability flag sequence includes capability satisfaction flags, condition dependency flags, and stability order indexes; and the WebXR application code entities include directly executable code segments, conditional judgment code segments, and sequential combination scripts.

3. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 1, characterized in that: The semantic parsing module includes: The input acquisition submodule acquires the text or voice content entered by the user in the WebXR application generation interface, reads the text character stream and speech transcription sequence sentence by sentence, detects the start and end positions of each sentence and records the sentence order index, collects the sentence length value and word count value, completes the order marking based on the sentence order index, and generates the sentence order mark quantity. Based on the sentence order markers, the semantic recognition submodule performs word comparison and label judgment on each sentence content. It matches and counts the words associated with spatial objects with the preset object vocabulary, matches and counts the words associated with interactive actions with the action instruction table, and matches and counts the words associated with input methods with the operation identifier table. The three types of matching values ​​are obtained and merged to obtain the semantic classification count value. The sequential integration submodule concatenates the spatial object description information, action description information, and operation description information in the original order of appearance according to the semantic classification count value and the corresponding sentence order index. It then assigns numbers to the description codes and forms a sequence mapping to generate a WebXR scene semantic description set.

4. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 3, characterized in that: The scene constraint module includes: The semantic inspection submodule obtains the set of semantic descriptions of the WebXR scene, checks each semantic description one by one, reads the spatial object description information marker value and the interactive behavior description information marker value in the description, judges the two types of marker values ​​within the same semantic description, identifies the semantic item sequence number that meets the judgment condition based on the co-occurrence judgment benchmark value, and generates a co-occurrence judgment sequence number. Based on the co-occurrence judgment sequence number, the relationship recording submodule calls the corresponding semantic description content, extracts the spatial object identifier value and the interaction behavior identifier value, pairs the two types of identifier values ​​and records the mapping number, establishes an object and interaction pairing record for each identified semantic description, and obtains the object interaction corresponding quantity; The sequential summary submodule arranges the corresponding object interaction quantities according to the original order of appearance of the semantic description, arranges the paired records in sequence to form a continuous sequence, and summarizes the sequence by number to generate a WebXR scene interaction constraint set.

5. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 4, characterized in that: In the WebXR scene semantic description set obtained by the semantic inspection submodule, each semantic description contains at least one spatial object description information marker value and at least one interactive behavior description information marker value. The co-occurrence judgment benchmark value is limited to a threshold value used to determine the number of spatial object description information marker values ​​and interactive behavior description information marker values ​​that appear simultaneously within the same semantic description. The number threshold value is not less than one. The process of identifying the semantic entry number that meets the judgment condition based on the co-occurrence judgment benchmark value is specifically limited to recording the original sequence number of the corresponding semantic description in the WebXR scene semantic description set only when the number of spatial object description information marker values ​​and the number of interactive behavior description information marker values ​​detected within the same semantic description both meet the number threshold value. The process of pairing the two types of marker values ​​and recording the mapping number is specifically limited to generating a one-to-one pairing relationship in the order of spatial object marker value first and interactive behavior marker value last, and assigning a unique incremental mapping number to each pairing relationship. The process of arranging according to the original order of appearance of the semantic description is specifically limited to sorting according to the original sequence number from small to large, and connecting the sorted pairing records sequentially to form a continuous number sequence.

6. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 4, characterized in that: The execution structure module includes: The node generation submodule obtains the WebXR scene interaction constraint set, reads the corresponding records of object identifier value and interaction identifier value one by one, generates an execution node number for each set of corresponding records, records the combination relationship between object number and interaction number, detects the continuity of number and marks the node position index, completes the node entry registration according to the node position index, and generates the execution node number quantity. Based on the execution node number, the condition supplementation submodule calls the corresponding object number and interaction number, collects the interaction trigger source identifier value, the state maintenance condition identifier value and the resource usage identifier value, merges the three types of identifier values ​​at the node level and records the attribute sequence number, and forms an attribute entry set for each node to obtain the node attribute combination quantity; The sequential connection submodule connects the execution nodes one after the other according to the original order of the interaction constraint set based on the combination of node attributes and the node position index. It records the connection number and sequence mapping value between nodes, forms a chain sequence of consecutive nodes and summarizes the numbers, and establishes a WebXR execution structure description sequence.

7. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 6, characterized in that: The stability determination module includes: The capability acquisition submodule calls the WebXR execution structure description sequence to obtain the runtime capability information provided by the browser environment and the runtime capability information provided by the WebXR session. It collects the number of computing units, the maximum memory usage, the graphics support level, and the session permission identifier. It standardizes the format and registers the capability values, establishes a capability item index table, and generates a quantity of runtime capability parameters. Based on the aforementioned operational capability parameters, the node determination submodule reads the resource usage requirements and interaction characteristic requirements for each execution node, compares the node resource requirement value with the corresponding capability upper limit value, marks records whose node requirement value does not exceed the capability upper limit value as directly runnable, and marks records whose node requirement value exceeds the capability upper limit value as conditionally dependent, and summarizes the node determination identifiers to obtain the node stability identifier quantity. The sequence arrangement submodule arranges the execution node determination identifiers according to the original connection order based on the node stability identifier quantity and the execution structure description sequence node sequence number, maps the stability identifiers and node numbers sequentially and records them to generate an execution stability mark sequence.

8. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 7, characterized in that: The number of computing units is limited to the minimum number of logical processing units that can be allocated simultaneously in the browser environment and the WebXR session. The maximum memory usage is limited to the minimum available memory capacity identified in the runtime capability information provided by the browser environment and the runtime capability information provided by the WebXR session. The process of standardizing the format and registering the capability values ​​is specifically limited to converting the number of computing units, the maximum memory usage, the graphics support level, and the session permission identifier into a unified data type and writing them into the capability entry index table according to a preset field order. The process of comparing the node resource requirement value with the corresponding capability limit value is specifically limited to comparing the resource usage requirement value with the computing unit quantity value, the memory usage limit value, and the graphics support level value one by one. When any comparison result exceeds the corresponding capability limit value, the condition dependency flag is recorded. The process of sequentially mapping and combining stability identifiers with node numbers is specifically defined as using the sequential number of the nodes in the execution structure description sequence as the primary key, and writing the corresponding node stability identifiers into the execution stability tag sequence in a one-to-one correspondence.

9. The intelligent generation system for WebXR immersive applications based on a large language model according to claim 7, characterized in that: The code generation module includes: The node mapping submodule reads the execution node number and stability identifier value one by one according to the execution stability marker sequence and the WebXR execution structure description sequence. For each execution node, it calls the object identifier, interaction identifier and node attribute entry, converts the node attribute entry into the corresponding interaction instruction parameter set, records the instruction sequence number and parameter length value, and generates the node instruction mapping quantity. Based on the node instruction mapping quantity and execution stability flag value, the condition generation submodule determines whether the node identifier is directly runnable or conditionally dependent. For directly runnable nodes, it extracts the instruction parameter set to form a code fragment number. For conditionally dependent nodes, it overlays the running condition judgment flag and generates a conditional code fragment number. The two types of code fragments are numbered and registered to obtain the code fragment identifier quantity. The sequential combination submodule concatenates the code fragments in sequence according to the code fragment identifier and the sequential number of the execution structure description sequence node, records the fragment connection order value and the total length count value, summarizes the overall combination result with a number, and outputs the WebXR application code entity.

10. A method for intelligent generation of WebXR immersive applications based on a large language model, characterized in that, The intelligent generation system for WebXR immersive applications based on a large language model, as described in any one of claims 1-9, is executed. Includes the following steps: S1: Obtain the text or voice content entered by the user in the WebXR application generation interface, read the input content sentence by sentence, identify the spatial object description information used to describe the WebXR scene, the action description information used to describe the XR interaction behavior, and the operation description information used to describe the input method, and organize the description information according to the original order of appearance to form a WebXR scene semantic description set. S2: Based on the WebXR scene semantic description set, check each semantic description to determine whether spatial object description information and interactive behavior description information appear simultaneously in the semantic description. For semantic descriptions that meet the judgment conditions, record the correspondence between the object and the interaction, and summarize the correspondence in the order of description to form a WebXR scene interaction constraint set. S3: Based on the WebXR scene interaction constraint set, generate an execution node description for each group of objects and interaction correspondences, supplement the execution node with the interaction trigger source, state maintenance conditions and resource usage requirements, and connect the execution nodes in the order of the interaction constraint set to establish a WebXR execution structure description sequence; S4: Call the WebXR execution structure description sequence to obtain the running capability information provided by the browser environment and WebXR session, judge the resource usage requirements and interaction characteristic requirements described by each execution node, distinguish between execution nodes that meet the running capability conditions and execution nodes that depend on the target running conditions, and organize the judgment results according to the execution structure order to generate an execution stability mark sequence. S5: Based on the execution stability marker sequence and the WebXR execution structure description sequence, the execution nodes are mapped to the corresponding WebXR interactive code content. For execution nodes that meet the running capability conditions, code segments that can be executed directly are generated. For execution nodes that depend on the target running conditions, code segments with running condition judgments are generated. The code segments are combined in the execution order to output the WebXR application code entity.

Images