A dynamic sign language dictionary adaptive updating method and system combined with reinforcement learning

Abstract

The present invention discloses a method and system for adaptively updating a dynamic sign language dictionary combined with reinforcement learning. The method identifies update needs by real-time monitoring of user search failure data, constructs entries by combining sign language linguistic rules with multimedia dynamic demonstration, and utilizes user feedback to drive a reinforcement learning model to continuously optimize entry quality. The method converts user behavior data into reinforcement learning reward signals, driving the model to dynamically adjust the semantic relevance and expression accuracy of the entries. The method can ensure that the new entries conform to the sign language grammar system through associative word network analysis. The multimedia dynamic demonstration makes up for the defects of static expression of traditional dictionaries, making the update process both user-oriented and linguistically standardized, avoiding the problems of delayed update, user feedback gaps, and static presentation limitations of traditional sign language dictionaries. The method achieves continuous self-optimization of entry quality through reinforcement learning, solving the problem of imbalance between update rate and update effect in sign language dictionary update methods in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of database structure technology, in particular to a database management technology, and more particularly to a method and system for adaptively updating a dynamic sign language dictionary combined with reinforcement learning. Background Art

[0002] Sign language is a visual, natural language that primarily conveys information through gestures, facial expressions, body movements, and changes in spatial orientation. It is not a simple translation of spoken language, but rather a complete language with its own unique grammatical rules and vocabulary. As the native language of the hearing-impaired, sign language is not only a core tool for daily communication but is also widely used in professional fields such as education, healthcare, and law to ensure equal participation for the deaf. Furthermore, technological advancements, the widespread use of sign language interpretation technology and accessible services are driving society's progress from eliminating communication barriers to deeper inclusion and cultural recognition for the deaf.

[0003] Sign language generates new words as society progresses. If traditional dictionaries are not updated promptly, there will be a gap in expression. Electronic dictionaries need to support features such as video presentations and AI searches, while paper versions struggle to accommodate dynamic language features, and delayed updates will impact the learning experience. Therefore, sign language requires a faster update frequency, but the traditional compilation process requires expert review and field research, resulting in delayed inclusion of new words. During the entry display process, customer feedback can increase the time required to modify new words, slowing down sign language dictionary updates and creating an imbalance between the effectiveness and efficiency of sign language dictionary updates.

[0004] Therefore, it is necessary to improve the sign language dictionary updating method in daily life to solve the above defects. Summary of the Invention

[0005] The present invention overcomes the shortcomings of the prior art and provides a method and system for adaptively updating a dynamic sign language dictionary combined with reinforcement learning, thereby solving the problem of imbalance between update rate and update effect in the prior art sign language dictionary update method.

[0006] To achieve the above objectives, the present invention adopts a technical solution: a method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning, comprising:

[0007] S1. Obtain the number of unsuccessful searches and the search frequency of a user's search for a word, and determine whether the word needs to be updated based on the number of unsuccessful searches and the search frequency; if it is determined that the word needs to be updated, issue an update instruction; if it is determined that the word does not need to be updated, mark the word as pending update, and the pending update mark is used to issue an update instruction for the word the next time it is determined;

[0008] S2. Collecting sign language linguistic processing information for the vocabulary to be updated, the sign language linguistic processing information including gesture parameter information and associated word information, collecting multimedia information based on the sign language linguistic processing information, and generating entries from the multimedia information;

[0009] S3. Collecting user term usage information, including positive usage information and negative usage information, performing reinforcement learning analysis based on the user term usage information, and determining the completion indicator based on the reinforcement learning analysis results;

[0010] S4. If the reinforcement learning analysis result is higher than the completion index, an update completion instruction is output; if the reinforcement learning analysis result is lower than the completion index, an update modification instruction is output, wherein the update modification instruction is used to modify the sign language linguistic processing information and the multimedia information, and perform reinforcement learning analysis on the modified sign language linguistic processing information and multimedia information until the reinforcement learning analysis result is higher than the completion index.

[0011] In a preferred embodiment of the present invention, the number of unsuccessful attempts refers to the number of times the user's search terms cannot be matched with the dictionary vocabulary, and the selection of the judgment terms that need to be updated is based on the number of unsuccessful attempts and the search frequency ranking, specifically the terms whose unsuccessful attempt ranking is within the number range and whose search frequency ranking is within the frequency range.

[0012] In a preferred embodiment of the present invention, the criteria for determining whether the vocabulary needs to be updated are: one or more of the number of unsuccessful attempts is higher than the unsuccessful attempt standard value and the search frequency is higher than the search frequency standard value, and the unsuccessful attempt standard value and the search frequency standard value are preset values ​​respectively. When the number of unsuccessful attempts is lower than the unsuccessful attempt standard value and the search frequency is lower than the search frequency standard value, it is determined that the vocabulary does not need to be updated.

[0013] In a preferred embodiment of the present invention, the gesture parameter information includes hand feature parameters, hand movement parameters, facial parameters and torso movement parameters, the associated word information includes semantic field association information, hyponym and hyponym information and synonym and antonym information, the multimedia information includes video, three-dimensional animation and pictures, and the process of making the multimedia information into entries is to place the multimedia information on the software interface or web page according to the layout.

[0014] In a preferred embodiment of the present invention, the completion index is the sum of the field average index and the trend prediction index, the field average index is the average value of the completion indexes of other words in the related field, and the trend prediction index is determined by the gesture parameter information rating.

[0015] In a preferred embodiment of the present invention, the reinforcement learning analysis is to assign points to the friendly usage information and the negative usage information respectively, the friendly usage information is assigned a positive value, and the negative usage information is assigned a negative value. The reinforcement learning analysis result is the sum of the points assigned to the friendly usage information and the negative usage information, and the completion indicator is a positive value.

[0016] In a preferred embodiment of the present invention, the usage-friendly information includes: gesture learning completion rate, average dwell time and associated word jump rate, and the usage-negative information includes: gesture complexity evaluation, movement clarity evaluation and semantic understanding difficulty evaluation. The gesture learning completion rate, the average dwell time, the associated word jump rate, the gesture complexity evaluation, the movement clarity evaluation and the semantic understanding difficulty evaluation are scored respectively.

[0017] In a preferred embodiment of the present invention, when the reinforcement learning analysis result is lower than the completion index and is a positive value, the associated word information is modified; when the reinforcement learning analysis result is lower than the completion index and is a negative value, the gesture parameter information and the associated word information are modified.

[0018] In a preferred embodiment of the present invention, the gesture parameter information is modified by increasing the movement amplitude and reducing the movement difficulty, and the associated word information is modified by increasing the number of associated words.

[0019] To achieve the above objectives, the second technical solution adopted by the present invention is: a dynamic sign language dictionary adaptive updating system combined with reinforcement learning, comprising:

[0020] A vocabulary acquisition unit, used to acquire the number of unsuccessful searches and search frequency of a user's vocabulary;

[0021] A fetching unit, used to fetch words to determine whether they are words that need to be updated;

[0022] A vocabulary update judgment unit, used to judge whether the vocabulary is a vocabulary that needs to be updated;

[0023] An update judgment instruction unit is used to issue an update instruction to the user or mark it as pending update;

[0024] a linguistic processing unit for performing sign language linguistic processing on the updated vocabulary;

[0025] Multimedia unit, used for collecting multimedia information and forming entries;

[0026] User usage unit, used to collect user entry usage information;

[0027] A database for developing and storing completion indicators for all vocabulary;

[0028] A reinforcement learning analysis unit is used to perform reinforcement learning processing on the vocabulary and obtain reinforcement learning analysis results;

[0029] a completion indicator judgment unit, connected to the database and the reinforcement learning analysis unit, respectively, for judging the relationship between the reinforcement learning result and the completion indicator;

[0030] An update completion unit, used to output an update completion instruction or an update modification instruction;

[0031] A modification unit is used to modify the sign language linguistic processing information and the multimedia information, and is connected to the reinforcement learning analysis unit.

[0032] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0033] (1) The present invention provides a dynamic sign language dictionary adaptive update method combined with reinforcement learning. It identifies the update needs by real-time monitoring of user search failure data, constructs entries by combining sign language linguistic rules with multimedia dynamic demonstration, and uses user feedback to drive the reinforcement learning model to continuously optimize the entry quality. It converts user behavior data into reinforcement learning reward signals to drive the model to dynamically adjust the semantic relevance and expression accuracy of the entry. Compared with the sign language dictionary update method in the prior art, it can ensure that the new entry conforms to the sign language grammar system through the network analysis of associated words. The multimedia dynamic demonstration makes up for the defects of the static expression of the traditional dictionary, making the update process both user-oriented and language-standardized, avoiding the problems of delayed update, user feedback gap, and static presentation limitations of the traditional sign language dictionary. Through reinforcement learning, it realizes continuous self-optimization of the entry quality, and solves the problem of imbalance between update rate and update effect in the sign language dictionary update method in the prior art.

[0034] (2) In the present invention, the criteria for determining whether a word needs to be updated are: one or more of the following: the number of unsuccessful attempts is higher than the unsuccessful attempt standard value and the search frequency is higher than the search frequency standard value. By setting preset standard values ​​for the number of unsuccessful attempts and the search frequency, the words that need to be updated are screened out. Compared with the existing technology, the updating of low-frequency or no-demand words can be avoided, ensuring that resources are concentrated on the words that users need most, thereby improving the updating effect.

[0035] (3) In the present invention, the completion index is the sum of the field average index and the trend prediction index. The field average index serves as a dynamic benchmark to ensure that the quality of each updated entry is not lower than the field average level. The trend prediction index is based on gesture parameter ratings, giving priority to updating high-demand or high-potential vocabulary. Compared with the existing technology, it avoids sacrificing quality for the pursuit of speed. At the same time, by referring to the average performance of similar vocabulary, the basic quality bottom line of the update is maintained.

[0036] (4) In the present invention, the comprehensive scoring mechanism of reinforcement learning converts user behavior into quantifiable reward signals, enabling the system to evaluate the quality of entries in real time. Compared with the existing technology, it can automatically adjust the update order according to the comprehensive score, quickly complete the update of high-positive-scoring entries, and enter the iterative modification of low-scoring entries, thus avoiding the contradiction between rate and effect caused by traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive efforts.

[0038] Figure 1 It is a method step diagram of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0041] Existing sign language dictionary update strategies mainly rely on manually preset static rules to determine priorities, such as time or classification order. This leads to delayed responses for high-demand vocabulary, such as hot words for emergencies, while low-demand vocabulary takes up limited resources. The process is also one-way and open-loop, lacking an effect tracking mechanism. As a result, updates that do not meet the standards are directly entered into the dictionary, requiring repeated revisions later, resulting in an inefficient cycle.

[0042] In response to the above technical problems, the concept of the present invention is to propose a dynamic sign language dictionary adaptive update method combined with reinforcement learning. It identifies update needs by real-time monitoring of user search failure data, constructs entries by combining sign language linguistic rules with multimedia dynamic demonstrations, and uses user feedback to drive the reinforcement learning model to continuously optimize the entry quality. It converts user behavior data into reinforcement learning reward signals, drives the model to dynamically adjust the semantic relevance and expression accuracy of the entries, and can ensure that the new entries conform to the sign language grammar system through associative word network analysis. Multimedia dynamic demonstration makes up for the defects of static expression of traditional dictionaries, so that the update process is both user-oriented and linguistically standardized, avoiding the problems of delayed updates, user feedback gaps, and static presentation limitations of traditional sign language dictionaries. Through reinforcement learning, continuous self-optimization of entry quality is achieved, solving the problem of imbalance between update rate and update effect in sign language dictionary update methods in the existing technology.

[0043] Exemplary Methods

[0044] Figure 1 The figure shows a flow chart of a method for adaptively updating a dynamic sign language dictionary in combination with reinforcement learning according to an embodiment of the present application. The updating method steps include:

[0045] S1. Obtain the number of unsuccessful searches and the search frequency of a user's search for a word, and determine whether the word needs to be updated based on the number of unsuccessful searches and the search frequency; if it is determined that the word needs to be updated, issue an update instruction; if it is determined that the word does not need to be updated, mark the word as pending update, and the pending update mark is used to issue an update instruction for the word in the next judgment;

[0046] S2. collecting sign language linguistic processing information for the vocabulary to be updated, the sign language linguistic processing information including gesture parameter information and associated word information, collecting multimedia information based on the sign language linguistic processing information, and creating entries from the multimedia information;

[0047] S3. Collect user term usage information, including positive usage information and negative usage information, conduct reinforcement learning analysis based on the user term usage information, and determine the completion indicator based on the reinforcement learning analysis results;

[0048] S4. If the reinforcement learning analysis result is higher than the completion index, an update completion instruction is output; if the reinforcement learning analysis result is lower than the completion index, an update modification instruction is output. The update modification instruction is used to modify the sign language linguistic processing information and multimedia information. The modified sign language linguistic processing information and multimedia information are subjected to reinforcement learning analysis until the reinforcement learning analysis result is higher than the completion index. The combination of the two makes the update strategy both forward-looking and robust, avoiding blind pursuit of speed or excessive conservatism.

[0049] Below, each step will be described in detail.

[0050] In step S1, the system monitors user search behavior in real time, counting the number of unsuccessful matches and search frequency. Unmarked words enter a pool for update, where their priority increases over time or as the number of failed matches accumulates, avoiding response delays caused by static rules.

[0051] In step S2, the system generates gesture parameters and a network of associated words for the vocabulary to be updated. This linguistic information is then coupled with dynamic visual information to form multimedia entries. For example, a complex gesture like "blockchain" is broken down into a step-by-step action video with keyframe parameters annotated, ensuring users can learn from multiple perspectives.

[0052] In step S3, after the entry is launched, the system tracks the user's use of friendly and negative information. This data is fed into the reinforcement learning model, which dynamically adjusts its strategy through a reward mechanism, such as +5 points for highly friendly gestures and -3 points for ambiguous gestures. For example, if a user gives up due to difficulty in spatially locating a "virtual reality" gesture, the model will infer that the range of motion should be simplified and generate a new solution.

[0053] In step S4, the reinforcement learning results must meet completion criteria. If they do, the entry is officially stored. If not, the system generates a modification instruction, triggering adjustments to the gesture parameters. The adjusted version undergoes reinforcement learning evaluation again, forming a "modify-test-iterate" cycle until the criteria are met.

[0054] The following is a detailed description of computer security management methods:

[0055] The number of unsuccessful attempts refers to the number of times a user's search terms cannot be matched with the dictionary vocabulary. The selection of words that need to be updated is based on the number of unsuccessful attempts and the search frequency ranking, specifically the words whose unsuccessful attempts rank within the number range and whose search frequency ranking is within the frequency range.

[0056] The number of unsuccessful searches reflects the frequency of user search failures, directly indicating a lack of desired vocabulary in the dictionary. Search frequency reflects user demand for a particular vocabulary, screening out frequently requested vocabulary. This dual-metric screening ensures that updated vocabulary is truly needed and frequently used by users, avoiding blind updates or updates to low-frequency vocabulary.

[0057] The number of unsuccessful searches solves the problem of slow update rates: by monitoring the number of unsuccessful searches in real time, it is possible to quickly identify words that need to be updated, thus preventing users from being unsatisfied for a long time. The search frequency solves the problem of poor update effectiveness: by prioritizing updates to frequently requested words, it ensures that the updated words are of maximum value to the user group. The combination of the two avoids the waste of resources caused by a too-fast update rate or the degradation of user experience caused by a too-slow update effect. Words with a high number of unsuccessful searches are updated first, addressing the urgent needs of users. Words that rank high in the search results are updated first, ensuring that frequently requested words are optimized. This dynamic adjustment mechanism makes the update process more flexible and can be optimized in real time based on user behavior.

[0058] The criteria for determining whether the word needs to be updated are: one or more of the following: the number of unsuccessful attempts is higher than the unsuccessful attempt standard value and the search frequency is higher than the search frequency standard value. The unsuccessful attempt standard value and the search frequency standard value are preset values ​​respectively. When the number of unsuccessful attempts is lower than the unsuccessful attempt standard value and the search frequency is lower than the search frequency standard value, it is determined that the word does not need to be updated.

[0059] A search frequency higher than the standard value allows users to quickly identify search terms that failed, directly resolving the issue of users not being able to find the terms they need and preventing them from experiencing long-term frustration. A search frequency higher than the standard value allows users to prioritize frequently searched terms, ensuring that frequently requested terms are updated promptly.

[0060] By setting preset standards for the number of unsuccessful searches and search frequency, we can filter out the words that truly need to be updated, avoiding updates to low-frequency or unneeded words. This targeted update ensures that resources are focused on the words that users need most, improving update effectiveness.

[0061] Determining updated vocabulary based on the number of unsuccessful searches can address slow update rates and provide a quicker response to user search failures. Determining updated vocabulary based on search frequency can address poor update effectiveness, ensuring that the updated vocabulary maximizes its value to the user base. The combination of these two approaches avoids resource waste caused by excessively fast update rates or a diminished user experience caused by slow update effectiveness. By filtering vocabulary for update using preset criteria, unnecessary updates to vocabulary with low unsuccessful searches and low search frequency are avoided, ensuring an efficient and targeted update process and minimizing resource waste.

[0062] Gesture parameter information includes hand feature parameters, hand movement parameters, facial parameters and torso movement parameters; associated word information includes semantic field association information, hyponym and hyponym information and synonym and antonym information; multimedia information includes videos, three-dimensional animations and pictures; and the process of making multimedia information into entries is to place the multimedia information on the software interface or web page according to the layout.

[0063] Detailed descriptions of hand feature parameters, hand movement parameters, facial parameters, and torso movement parameters enable more accurate representation of sign language movements, ensuring that the semantic expression of entries conforms to the sign language grammar system. By using semantic field association information, hyponym and hyponym information, and synonym and antonym information, the semantic relevance of vocabulary is expanded, ensuring that the expression of entries is more comprehensive and better meets user needs. Dynamic demonstrations through videos, 3D animations, and images overcome the limitations of static expression in traditional dictionaries, improving the intuitiveness and usability of entries.

[0064] The completion index is the sum of the domain average index and the trend prediction index. The domain average index is the average of the completion indexes of other vocabularies in the relevant domain. The trend prediction index is determined by the gesture parameter rating. Based on the average completion index of other vocabularies in the relevant domain, it provides an objective baseline to ensure the stability and consistency of entry updates. The domain average index serves as a dynamic benchmark, ensuring that the quality of each entry update is at least as high as the domain average. This avoids sacrificing quality for speed while maintaining a basic quality baseline for updates by referencing the average performance of similar vocabularies. The trend prediction index, based on gesture parameter ratings, prioritizes updates for high-demand or high-potential vocabularies. By predicting future usage trends, the system proactively updates vocabularies that are likely to be used frequently, accelerating response time and ensuring that updates align with user needs. The completion index is a positive value, calculated by rounding the sum of the domain average index and the trend prediction index.

[0065] Trend predictions within completion metrics guide the reinforcement learning model to focus on high-value terms, driving the system to prioritize resource optimization for these terms. Domain averages serve as quality anchors, ensuring the optimization process adheres to linguistic norms. This combination of factors makes the update strategy both forward-looking and robust, avoiding either blind pursuit of speed or excessive conservatism. While traditional methods can lead to rigid updates due to fixed thresholds, this approach flexibly adjusts update priorities through dynamic metrics.

[0066] Reinforcement learning analysis is to assign points to the use of friendly information and the use of negative information respectively. The use of friendly information is assigned a positive value, and the use of negative information is assigned a negative value. The reinforcement learning analysis result is the sum of the points assigned to the use of friendly information and the use of negative information, and the completion indicator is a positive value.

[0067] The comprehensive scoring mechanism converts user behavior into quantifiable reward signals, enabling the system to evaluate entry quality in real time. The update order is automatically adjusted based on the comprehensive score, with high-scoring entries quickly updated and low-scoring entries undergoing iterative revisions. This avoids the conflict between speed and effectiveness caused by the traditional "fixed-order update" approach.

[0068] Friendly information used includes: gesture learning completion rate, average dwell time and associated word jump rate; negative information used includes: gesture complexity evaluation, movement clarity evaluation and semantic understanding difficulty evaluation. Gesture learning completion rate, average dwell time, associated word jump rate, gesture complexity evaluation, movement clarity evaluation and semantic understanding difficulty evaluation are scored respectively, and the scoring range is 1-5 points.

[0069] The gesture learning completion rate refers to the gesture learning completion rate after the user views the entry information for the first time. The user is asked whether he or she has learned the gesture through a questionnaire or other form. The scores are 10 points, 8 points, 6 points, 4 points, and 2 points according to the cases of ≥85%, [70%, 80%, [50%, 70%, [30%, 50%), and less than 30%, respectively.

[0070] The average dwell time refers to the time users spend on an entry page, reflecting the effectiveness and learning value of the content. It is scored as 5 points, 4 points, 3 points, 2 points, and 1 point for ≥1 minute, [40s, 1 minute], [25s, 40s], [10s, 25s], and less than 10 seconds, respectively.

[0071] The jump rate of related words refers to the proportion of users clicking on related recommended terms, and is scored as 5 points, 4 points, 3 points, 2 points and 1 point according to the cases of ≥40%, [30%, 40%), [20%, 30%), [10%, 20%) and less than 10% respectively.

[0072] The gesture complexity evaluation is specifically based on the proportion of users who rated it as complex. Scores are assigned as -10 points, -8 points, -6 points, -4 points, and -2 points for situations where the percentage is ≥60%, [40%, 60%), [25%, 40%), [10%, 25%), and less than 10%, respectively.

[0073] The motion clarity evaluation is specifically the proportion of evaluated users who rated the motion as unclear. The scores are -5 points, -4 points, -3 points, -2 points, and -1 points according to the cases of ≥50%, [35%, 50%), [20%, 35%), [10%, 20%), and less than 10%, respectively.

[0074] The semantic understanding difficulty evaluation is specifically the proportion of users who rated the semantics as difficult to understand among the evaluated users. The scores are -5 points, -4 points, -3 points, -2 points, and -1 points according to the cases of ≥50%, [35%, 50%), [20%, 35%), [10%, 20%), and less than 10%, respectively.

[0075] Vocabulary with high gesture learning completion rates reflects clear user needs, allowing the system to quickly prioritize updates and improve update rates. A high average dwell time indicates engaging content, allowing the system to retain its core design and fine-tune interaction details, balancing speed and quality. A high jump rate for related words indicates a well-structured knowledge network, allowing the system to accelerate the expansion of related vocabulary links and improve user learning efficiency.

[0076] Vocabulary with low gesture complexity ratings should prioritize simplifying the action parameters to prevent users from giving up due to learning difficulties. Vocabulary with poor action clarity ratings should have the action range increased or step-by-step demonstrations added to ensure the updated effect meets the standards.

[0077] When the reinforcement learning analysis result is lower than the completion index and is a positive value, the associated word information is modified; when the reinforcement learning analysis result is lower than the completion index and is a negative value, the gesture parameter information and the associated word information are modified.

[0078] If the analysis result falls below the target but is positive, such as high user acceptance but slightly complex gestures, the system prioritizes adjustments to associated word information, such as adding synonyms. If the result is negative, such as difficult gestures and ambiguous semantics, both gesture parameters and associated words are modified. This hierarchical response mechanism avoids over-adjustments while ensuring critical defects are addressed. By identifying deficiencies in new word acceptance, the system analyzes whether the new word is difficult to understand from a sign language gesture design perspective or semantically, and makes targeted modifications accordingly.

[0079] Gesture parameter information can be modified by increasing the amplitude and reducing the difficulty of the gesture, while associated word information can be modified by increasing the number of associated words. Simplifying complex gestures can speed up user learning and reduce repeated queries caused by difficult-to-master gestures. The system eliminates the need to design multiple alternatives for complex gestures, directly optimizing core motion parameters and increasing update rates. Increasing the amplitude of the gesture enhances visualization and helps users quickly identify the meaning of the gesture. By adjusting the video angle or adding dynamic markers, the system can improve clarity and accelerate the update process without redesigning the entire gesture set.

[0080] Add multiple synonyms or contextual associations to a single gesture to form a semantic network. This allows users to access the target gesture through different vocabulary, reducing the difficulty of understanding and query failures caused by inaccurate single associations, indirectly improving update effectiveness. For high-frequency but low-complexity words, simply adding associations can quickly complete updates and increase speed. For low-frequency but high-complexity words, prioritize reducing the difficulty of the gesture and then gradually expand the associations to ensure effectiveness.

[0081] Exemplary devices

[0082] To achieve the above objectives, the second technical solution adopted by the present invention is: a dynamic sign language dictionary adaptive updating system combined with reinforcement learning, comprising:

[0083] A vocabulary acquisition unit, used to acquire the number of unsuccessful searches and search frequency of a user's vocabulary;

[0084] A fetching unit, used to fetch words to determine whether they are words that need to be updated;

[0085] A vocabulary update judgment unit, used to judge whether the vocabulary is a vocabulary that needs to be updated;

[0086] An update judgment instruction unit is used to issue an update instruction to the user or mark it as pending update;

[0087] a linguistic processing unit for performing sign language linguistic processing on the updated vocabulary;

[0088] Multimedia unit, used for collecting multimedia information and forming entries;

[0089] User usage unit, used to collect user entry usage information;

[0090] A database for developing and storing completion indicators for all vocabulary;

[0091] A reinforcement learning analysis unit is used to perform reinforcement learning processing on the vocabulary and obtain reinforcement learning analysis results;

[0092] The completion indicator judgment unit is connected to the database and the reinforcement learning analysis unit respectively, and is used to judge the relationship between the reinforcement learning results and the completion indicators;

[0093] An update completion unit, used to output an update completion instruction or an update modification instruction;

[0094] The modification unit is used to modify the sign language linguistics processing information and multimedia information, and is connected to the reinforcement learning analysis unit.

[0095] Exemplary electronic devices

[0096] An electronic device, comprising:

[0097] Processor; a processor may be a central processing unit (CPU) or other form of processing unit having data processing capability and / or instruction execution capability.

[0098] A memory stores computer program instructions, which, when executed by a processor, enable the processor to execute a dynamic sign language dictionary adaptive updating method combined with reinforcement learning.

[0099] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory and / or cache memory. The non-volatile memory may include, for example, read-only memory, a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the dynamic sign language dictionary adaptive update method combined with reinforcement learning of the various embodiments of the present application described above and / or other desired functions.

[0100] A computer-readable storage medium stores computer program instructions, which, when executed by a processor, enable the processor to execute a dynamic sign language dictionary adaptive updating method combined with reinforcement learning.

[0101] Computer readable storage media can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, but is not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0102] The above description is based on the ideal embodiment of the present invention. Based on the above description, relevant personnel can make various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the content of the specification and must be determined according to the scope of the claims.

Claims

1. A method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning, characterized in that: include: S1. Obtain the number of unsuccessful searches and search frequency of a user's search terms, and determine whether the term needs to be updated based on the number of unsuccessful searches and search frequency; If it is determined that the vocabulary needs to be updated, an update instruction is issued; if it is determined that the vocabulary does not need to be updated, the vocabulary that does not need to be updated is marked as pending update, and the pending update mark is used to issue an update instruction to the vocabulary when it is determined next time; S2. Collecting sign language linguistic processing information for the vocabulary to be updated, the sign language linguistic processing information including gesture parameter information and associated word information, collecting multimedia information based on the sign language linguistic processing information, and generating entries from the multimedia information; S3. Collecting user term usage information, including positive usage information and negative usage information, performing reinforcement learning analysis based on the user term usage information, and determining the completion indicator based on the reinforcement learning analysis results; S4. If the reinforcement learning analysis result is higher than the completion index, output an update completion instruction; If the reinforcement learning analysis result is lower than the completion index, an update modification instruction is output, and the update modification instruction is used to modify the sign language linguistic processing information and the multimedia information, and reinforcement learning analysis is performed on the modified sign language linguistic processing information and multimedia information until the reinforcement learning analysis result is higher than the completion index.

2. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: The number of unsuccessful attempts refers to the number of times the user's search terms cannot be matched with the dictionary vocabulary. The selection of the vocabulary that needs to be updated is based on the number of unsuccessful attempts and the search frequency ranking, specifically the vocabulary whose unsuccessful attempt ranking is within the number range and whose search frequency ranking is within the frequency range.

3. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: The criteria for determining whether the word needs to be updated are: one or more of the following: the number of unsuccessful attempts is higher than the unsuccessful attempt standard value and the search frequency is higher than the search frequency standard value. The unsuccessful attempt standard value and the search frequency standard value are preset values ​​respectively. When the number of unsuccessful attempts is lower than the unsuccessful attempt standard value and the search frequency is lower than the search frequency standard value, it is determined that the word does not need to be updated.

4. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: The gesture parameter information includes hand feature parameters, hand movement parameters, facial parameters and torso movement parameters; the associated word information includes semantic field association information, hyponym and hyponym information and synonym and antonym information; the multimedia information includes video, three-dimensional animation and pictures; and the process of making the multimedia information into entries is to place the multimedia information on the software interface or web page according to the layout.

5. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: The completion index is the sum of the field average index and the trend prediction index. The field average index is the average value of the completion indexes of other words in the relevant field. The trend prediction index is determined by the gesture parameter information rating.

6. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: The reinforcement learning analysis is to assign points to the friendly usage information and the negative usage information respectively, the friendly usage information is assigned a positive value, and the negative usage information is assigned a negative value. The reinforcement learning analysis result is the sum of the points assigned to the friendly usage information and the negative usage information, and the completion index is a positive value.

7. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 6, characterized in that: The friendly usage information includes: gesture learning completion rate, average stay time and associated word jump rate; the negative usage information includes: gesture complexity evaluation, movement clarity evaluation and semantic understanding difficulty evaluation. The gesture learning completion rate, the average stay time, the associated word jump rate, the gesture complexity evaluation, the movement clarity evaluation and the semantic understanding difficulty evaluation are scored respectively.

8. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 1, characterized in that: When the reinforcement learning analysis result is lower than the completion index and is a positive value, the associated word information is modified; when the reinforcement learning analysis result is lower than the completion index and is a negative value, the gesture parameter information and the associated word information are modified.

9. The method for adaptively updating a dynamic sign language dictionary combined with reinforcement learning according to claim 8, characterized in that: The gesture parameter information is modified by increasing the movement amplitude and reducing the movement difficulty, and the associated word information is modified by increasing the number of associated words.

10. A dynamic sign language dictionary adaptive updating system combined with reinforcement learning, based on a dynamic sign language dictionary adaptive updating method combined with reinforcement learning according to any one of claims 1 to 9, characterized in that: include: A vocabulary acquisition unit, used to acquire the number of unsuccessful searches and search frequency of a user's vocabulary; A fetching unit, used to fetch words to determine whether they are words that need to be updated; A vocabulary update judgment unit, used to judge whether the vocabulary is a vocabulary that needs to be updated; An update judgment instruction unit is used to issue an update instruction to the user or mark it as pending update; a linguistic processing unit for performing sign language linguistic processing on the updated vocabulary; Multimedia unit, used for collecting multimedia information and forming entries; User usage unit, used to collect user entry usage information; A database for developing and storing completion indicators for all vocabulary; A reinforcement learning analysis unit is used to perform reinforcement learning processing on the vocabulary and obtain reinforcement learning analysis results; a completion indicator judgment unit, connected to the database and the reinforcement learning analysis unit, respectively, for judging the relationship between the reinforcement learning result and the completion indicator; An update completion unit, used to output an update completion instruction or an update modification instruction; A modification unit is used to modify the sign language linguistic processing information and the multimedia information, and is connected to the reinforcement learning analysis unit.

Images