Inquiry method of operation instructions, computer program product and query system related to computer program product

Abstract

The invention provides an operation instruction query method, a computer program product and a system related to the computer program product. Software programs are stored on the computer program product, and the software programs execute the query method of the operation instructions. The query method comprises the following steps: firstly, translating a non-English query character string into an English query character string by a language translator according to a language identifier; secondly, converting the English query character string into an English interaction prompt character string according to the at least one high-relevance symbol vector; wherein the high-relevance symbol vector is related to an operation instruction of the electronic device. Then, according to the language identifier, the language translator translates an English reply character string, which is inferred and generated based on the English interaction prompt character string, into a non-English reply character string.

Description

Technical Field

[0001] The present invention relates to a method for querying operation instructions, a computer program product, and a query platform related thereto, and particularly to a method for querying operation instructions, a computer program product, and a query platform related thereto that respond to the user's habitual language type. Background Art

[0002] In recent years, the method of using artificial intelligence (AI for short) for data query has become increasingly popular. Data that used to take a lot of time and effort to query can now be quickly found with the assistance of a generative pre-trained transformer model (GPT for short) based on transformers. For example, ChatGPT can be trained through deep learning with a large amount of data retrieved from the Internet and have conversations with users. Another example is that Copilot launched by Microsoft can assist users in dealing with more general or upper-layer application software usage problems. For example, it provides a draft function when paired with Outlook, and can write text when paired with Word.

[0003] Whether it is ChatGPT or Copilot, existing AI tools developed based on large language models (LLMs for short) are located in the cloud. For users, this means that the query will be affected by network bandwidth. Moreover, due to the wide data sources of current AI tools, if users want to ask more specific questions, these AI tools still cannot accurately provide or reply with more accurate information.

[0004] Currently, many electronic products have more and more functions, and users easily encounter situations where they want to operate a certain function but don't know where to start because they are not familiar with how to set the function. However, current AI tools can only provide relatively general information reply functions. When users ask about operation instructions related to the electronic products they are using, current AI tools may even give irrelevant answers (hallucination). Therefore, current AI tools still cannot assist users in quickly and accurately learning how to operate or set electronic devices. Summary of the Invention

[0005] The present invention relates to a method for querying operation instructions, a computer program product, and a query system related thereto. The query method applied to the query system has a translation function and an interaction function, and improves the interaction effect based on the English version of the user manual of the electronic device, so as to facilitate users to query the operation instructions of the electronic device.

[0006] According to a first aspect of the present invention, a method for querying operation instructions is provided. The query method includes the following steps: First, a language translator translates a non-English query string into an English query string according to a language identifier. Second, according to at least one highly relevant token vector, the English query string is converted into an English interactive prompt string. The highly relevant token vector is related to the operation instructions of the electronic device. Then, the language translator translates the English reply string inferred based on the English interactive prompt string into a non-English reply string according to the language identifier.

[0007] According to a second aspect of the present invention, a computer program product storing a software program is provided. The software program executes a method for querying operation instructions, and the query method includes the following steps: First, a language translator translates a non-English query string into an English query string according to a language identifier. Second, according to at least one highly relevant token vector, the English query string is converted into an English interactive prompt string. The highly relevant token vector is related to the operation instructions of the electronic device. Then, the language translator translates the English reply string inferred based on the English interactive prompt string into a non-English reply string according to the language identifier.

[0008] According to a third aspect of the present invention, a query system is provided. The query system includes: a query platform. The query platform includes: a language translator and a semantic core module. The language translator includes: an input translation model and an output translation model. The input translation model translates a non-English query string into an English query string according to a language identifier. The semantic core module converts the English query string into an English interactive prompt string according to a highly relevant token vector. The highly relevant token vector is related to the operation instructions of the electronic device. The output translation model translates the English reply string inferred based on the English interactive prompt string into a non-English reply string according to the language identifier.

[0009] For a better understanding of the above and other aspects of the present invention, the following specific embodiments are given, and are described in detail in conjunction with the accompanying drawings as follows. Description of the Drawings

[0010] Figure 1 is a block diagram of a query system conceived according to the present invention;

[0011] Figure 2 is a schematic diagram of setting the query platform in an electronic device;

[0012] Figure 3 is a schematic diagram of setting at least a part of the query platform in a server, in cooperation with an electronic device;

[0013] Figure 4 is a relational diagram of a query platform conceived according to the present invention for processing and converting materials; and Figure 5A 、 5B is a flowchart of a query method conceived according to the present invention.

[0014] Reference numerals:

[0015] 1: Query system

[0016] 18a: English version user manual

[0017] 18c: Syntax analyzer

[0018] 18e, Jfile: JSON file

[0019] 18g: Word vector model

[0020] 10: Query platform

[0021] 11: Language translator

[0022] 13: Semantic core module

[0023] 131: Vector database

[0024] 133: Hint message adjustment module

[0025] 15: Pretrained transformer

[0026] 18a: English version user manual

[0027] qryTXT(usrLNG): Non-English query string rTXT(usrLNG): Non-English response string

[0028] qryTXT(eng): English query string

[0029] rTXT(eng): English response string

[0030] pptTXT(eng): English interactive hint string

[0031] TOKvec: Pre-stored token vector

[0032] TOKvec_rel: High-correlation token vector

[0033] 20, 30: Electronic device

[0034] 201, 301: Microphone

[0035] 203, 303: Keyboard

[0036] 21a, 31a: Control module

[0037] 21c, 31c: Storage module

[0038] 205, 305: Speaker

[0039] 207, 307: Screen

[0040] 33: Network

[0041] 35: Server

[0042] usrLNG_ID: Language identifier

[0043] 111: Language recognizer

[0044] 113: Language translation model

[0045] 113a, intrMDL: Input translation model

[0046] 113c, otrMDL: Output translation model

[0047] S301, S303, S305, S307, S309, S311, S313a, S313c, S315, S317, S319, S321: Steps Detailed implementation manners

[0048] In order to assist a user in quickly and accurately querying operation instructions when the user is not familiar with the functions of an electronic product, the present invention provides a query system as shown in Figure 1 . The query system of the present invention generates an instruction in response to a query operation of the user, based on an English-language user manual of the electronic device operated by the user, so that the user can effectively obtain operation instructions related to the electronic device he operates. Furthermore, the query system of the present invention further provides a translation function, enabling a user who is not familiar with English to query operation instructions in his own familiar language (the user's habitual language usrLNG).

[0049] Please refer to Figure 1 , which is a block diagram of the query system conceived according to the present invention. The query system 1 includes a semantic parser 18c, a text vector model 18g, and a query platform 10. The semantic parser 18c and the text vector model 18g can be connected to each other through a signal connection or an electrical connection. The text vector model 18g is signal-connected to the query platform 10.

[0050] To assist users in understanding / familiarizing themselves with how to operate an electronic device, the manufacturer provides an English-language user manual 18a for the electronic device. In the English-language user manual 18a, instructions on how to perform various operations on the electronic device are described. Before the electronic device leaves the factory, the manufacturer uses a syntax analyzer 18c to pre-analyze the content of the English-language user manual 18a (various operation instructions related to the electronic device).

[0051] The results of the syntax analysis by the syntax analyzer 18c will be stored in the form of a JavaScript Object Notation (JSON) file (Jfile) 18e. Subsequently, the text vector model 18g further performs vector calculation and reformatting on the content of the JSON file (Jfile) 18e to generate multiple pre-stored token vectors (TOKvec) stored in the vector database 131. The query platform 10 can be fully built into the electronic device, fully set up on a server that is signal-connected to the electronic device, or partially set up on the electronic device and partially set up on the server.

[0052] The query platform 10 further includes: a language translator 11, a Semantic Kernel 13, and a Pre-trained Transformer 15. Among them, the Semantic Kernel 13 further includes a vector database 131 and a prompt tuning module 133.

[0053] The vector database 131 is used to provide functions for efficient storage and querying of the pre-stored token vectors TOKvec. The pre-stored token vectors TOKvec stored in the vector database 131 are high-dimensional vectors. Moreover, each pre-stored token vector TOKvec represents a feature or attribute of a certain data in the English-language user manual 18a. Before storing the pre-stored token vectors TOKvec in the vector database 131, the data in the JSON file (Jfile) 18e must be pre-processed using a vector format. Therefore, the JSON file (Jfile) 18e generated by the syntax analyzer 18c must be further used in conjunction with the text vector model 18g to convert the tokens in the JSON file (Jfile) 18e into pre-stored token vectors TOKvec.

[0054] The parser 18c internally includes: a tokenizer (Tokenizer) for splitting (textsplitter) the statements in the English version of the user manual 18a into smaller and more understandable tokens, an English thesaurus, and a small language model established using a neural network, etc.

[0055] Details on how the parser 18c decomposes statements and splits tokens, and how the text vector model 18g performs data vectorization processing will not be elaborated here. Generally, the process of token splitting and data vectorization for non-English languages is much more complex than that for English. Therefore, the query platform 10 of the present invention is provided with a language translator 11, enabling the vector database 131 to only consider the pre-stored token vectors TOKvec generated based on English. Accordingly, according to the parser 18c and the text vector model 18g conceived in the present invention, only the English version of the user manual 18a needs to be subjected to syntax analysis and data vectorization processing. Therefore, the processing speeds of the parser 18c and the text vector model 18g in this case are both relatively fast.

[0056] The vector database 131 can be used to record the vector values (Embedding) of the pre-stored token vectors TOKvec, the original text content in the JSON file (Jfile) 18e, and the number of tokens (Token) required for the pre-stored token vectors TOKvec, etc. Details on the operation of the text vector model 18g and the establishment method of the vector database 131, including how to establish a vector index, how to determine the fields within the vector index, the vector data type, the correlation algorithm between vectors, etc., will not be elaborated in this article. Briefly, after the data vectorization processing by the text vector model 18g, the pre-stored token vectors TOKvec stored in the vector database 131 are structured data, allowing the vector database 131 to perform similarity queries.

[0057] The pre-stored token vectors TOKvec stored in the vector database 131 are pre-established by the manufacturer before the electronic device leaves the factory. Moreover, the vector database 131 may be located in the cloud or stored in the storage module within the electronic device. Considering the storage space size, the source of the pre-stored token vectors TOKvec stored in the vector database 131 may only come from one English version of the user manual 18a. Or, considering data integrity, the source of the pre-stored token vectors TOKvec stored in the vector database 131 may come from multiple English versions of the user manual 18a. For the considerations in this part regarding applications, it can be selected according to actual requirements.

[0058] After the user starts using the electronic device and assuming that the user wants to perform some operations that the user is not familiar with on the electronic device, the user can use the query platform 10 to query the setting steps on how to operate the electronic device. Since the electronic device may be sold all over the world, when the user queries the operation instructions, the language that may be used is very likely not English. In this article, the query content input by the user is referred to as the non-English query string qryTXT(usrLNG).

[0059] The language translator 11 in the query platform 10 first translates the non-English query string qryTXT(usrLNG) into an English query string qryTXT(eng), and then transmits the English query string qryTXT(eng) to the prompt information adjustment module 133 and the vector database 131. Among them, the language translator 11, the information adjustment module 133, and the pre-trained transformer 15 can adopt different types and unequal numbers of inference accelerators. For example, the inference accelerator can be a neural-network processing unit (abbreviated as NPU), a graphics processing unit (abbreviated as GPU), a field programmable gate array (abbreviated as FPGA), etc.

[0060] According to the English query string qryTXT(eng), the vector database 131 further queries one or more highly relevant token vectors TOKvec_rel in the multiple pre-stored token vectors TOKvec stored in the vector database 131 that have a high correlation with the English query string qryTXT(eng).

[0061] According to the concept of the present invention, the vector database 131 can be further used in combination with a large language model (abbreviated as LLM). For example, steps such as semantic search, similarity search, and recommendation engine are performed on the pre-stored token vectors TOKvec to find at least one highly relevant token vector TOKvec_rel that is more in line with the operation instructions that the user wants to query. Moreover, the vector database 131 transmits the one or more highly relevant token vectors TOKvec_rel queried to the prompt tuning module 133.

[0062] When the vector database 131 retrieves multiple highly relevant token vectors TOKvec_rel, the vector database 131 can further sort these multiple highly relevant token vectors TOKvec_rel according to the degree of relevance. Moreover, based on the sorting result, the vector database 131 selects several (e.g., 3) highly relevant token vectors TOKvec_rel from them and transmits them to the prompt tuning module 133.

[0063] In actual application, while the language translator 11 transmits the non-English query string qryTXT(usrLNG), it can also attach the interpretation data (metadata) related to the model of the electronic device. In this way, when the vector database 131 receives the English query string qryTXT(eng) from the input translation model (intrMDL) 113a, it can further combine this interpretation data related to the model of the electronic device to find more accurate highly relevant token vectors TOKvec_rel. For example, the query requirements of users who purchase gaming computer models may be more advanced (e.g., tuning or optimizing computer performance), which are different from the query requirements of document operations of users who purchase basic computer models. If the language translator 11 transmits such interpretation data in the background, it can assist the vector database 131 in generating highly relevant token vectors TOKvec_rel more accurately.

[0064] As mentioned above, the prompt tuning module 133 receives the English query string qryTXT(eng) from the language translator 11; and, receives the highly relevant token vectors TOKvec_rel from the vector database 131. The prompt tuning module 133 is a conversion model for natural language processing (Natural Language Processing, abbreviated as NLP) based on machine learning technology. In short, the prompt tuning module 133 combines the English query string qryTXT(eng) with the highly relevant token vectors TOKvec_rel provided by the vector database 131 to convert the English query string qryTXT(eng) into an English interactive prompt string pptTXT(eng) that is easier for the pre-trained transformer 15 to understand.

[0065] Continuing from the above, the English interactive prompt string pptTXT(eng) can be regarded as the combination of the English query string qryTXT(eng) and the content of the highly relevant token vector TOKvec_rel. Also, because the highly relevant token vector TOKvec_rel is derived from the English version of the user manual 18a of the electronic device. Therefore, when describing the query conditions, the English interactive prompt string pptTXT(eng) is significantly closer to the actual situation of the electronic device than the English query string qryTXT(eng) generated by translation.

[0066] In actual application, the prompt information adjustment module 133 can transmit the English interactive prompt string pptTXT(eng) presented in the form of a question to the pre-trained transformer 15 based on methods such as Prefix-Tuning or Prompt Tuning. The pre-trained transformer 15 can be various large language models using generative artificial intelligence. For example, GPT3.5, GPT-3.5Turbo, GPT-4, Llama 2, etc.

[0067] The pre-trained transformer 15 makes inferences in response to the English interactive prompt string pptTXT(eng) and generates an English response string rTXT(eng) based on the inference results. After the pre-trained transformer 15 transmits the English response string rTXT(eng) to the language translator 11, the language translator 11 then translates the English response string rTXT(eng) into a non-English response string rTXT(usrLNG) for the user to refer to.

[0068] According to the concept of the present invention, the implementation method and the setting location of the query platform 10 can be determined according to the functions of the electronic device operated by the user and do not need to be limited. For example, when the electronic device operated by the user is a high-end product, it means that the processor speed of the electronic device is fast enough and the capacity of the storage module is large enough. For this kind of application, the manufacturer can set the entire query platform 10 in the electronic device, such as Figure 2 shown.

[0069] Please refer to Figure 2 , which is a schematic diagram of setting the query platform in the electronic device. The electronic device 20 includes a control module 21a, a storage module 21c, input devices (such as a microphone 201, a keyboard 203, etc.) and output devices (such as a speaker 205, a screen 207, etc.). In Figure 2In this case, it is assumed that the electronic device 20 belongs to a higher-order product. Therefore, the control module 21a has a faster processing speed, and the storage module 21c has a larger capacity, and can be directly used to execute the functions of the query platform 10. For example, the control module 21a may include one or more central processing units (CPUs), graphics processing units, or a combination thereof.

[0070] According to the concept of the present invention, the input method of the non-English query string qryTXT(usrLNG) received by the query platform 10 is not limited. For example, the user can input it by typing, handwriting, or voice input. Consequently, depending on the method of inputting the non-English query string qryTXT(usrLNG), the input device can be the keyboard 203, the touch panel, or the microphone 201. When it is the touch panel 203 or the microphone 201, the electronic device can further provide components such as a handwriting recognition module and a voice recognition module. Similarly, the method of outputting the non-English reply string rTXT(usrLNG) is not limited either. For example, the query platform 10 can be paired with a screen 207 for display, or played through a speaker 205. Whether it is displayed on the screen 207 or played through the speaker 205, the content replied by the query platform 10 has been pre-converted into the user's habitual language usrLNG for the user to understand. Regarding the types of input devices and output devices, it is not limited to the examples here.

[0071] In Figure 2 this case, the storage module 21c is used to store the vector database 131 and the database used by the pre-trained transformer 15, and the control module 21a is used to execute the functions of the language translator 11, the prompt message adjustment module 133, and the pre-trained transformer 15. When the electronic device adopts this architecture, it is necessary to pre-load the settings of the query platform 10 before starting to execute. In addition, adopting Figure 2 the architecture of the query platform 10 does not need to query through the network and does not require network bandwidth. Therefore, adopting Figure 2 the architecture of the query platform 10 requires a longer initial setup time, but the speed is faster during subsequent query operation instructions.

[0072] When the electronic device that the user intends to use for querying operation instructions belongs to a lower-order product, it means that the processor speed of the electronic device is slower and the capacity of the storage module is limited. For such applications, the manufacturer can set a part of the query platform 10 or the entire query platform 10 on a remote server. When the user wants to query the operation instructions of the electronic device, the electronic device accesses the server through the network 33.

[0073] Please refer to Figure 3, which is a schematic diagram of setting at least a part of the query platform on the server and matching with the electronic device. The electronic device 30 includes a control module 31a, a storage module 31c, a communication module 31e, an input device (such as a microphone 301, a keyboard 303, etc.) and an output device (such as a speaker 305, a screen 307, etc.). The control module 31a is electrically connected to the storage module 31c, the communication module 31e, the microphone 301, the keyboard 303, the speaker 305 and the screen 307. In addition, the communication module 31e and the server 35 set in the cloud are signal-connected through the network 33.

[0074] In Figure 3 , the control module 31a communicates with the external server 35 through the communication module 31e. Since the control module 31a only needs to set the online relationship with the server 35, the required initial setup time is relatively short. However, when subsequent operation instructions are queried, they all need to be transmitted and received through the network 33. Consequently, the query process requires network bandwidth, and when individual operation instructions are queried, the required query time is relatively long.

[0075] In actual application, the manufacturer can set part of the query platform 10 on the electronic device 30 and part of the query platform 10 in the cloud to reduce the dependence on network bandwidth. For example, one of the language translator 11, the semantic core module 13, and the pre-trained transformer 15 can be set on the electronic device, and the other two are set on the server 35. Or, two of the language translator 11, the semantic core module 13, and the pre-trained transformer 15 can be set on the electronic device 30, and the remaining one is set on the server 35.

[0076] Manufacturers of the electronic devices 20 and 30 can determine how to decide the configuration location of the internal components of the query platform 10 according to the general operation habits of users. For example, when the manufacturer judges that according to the general situation of users operating other software, the sum of the required processor and memory space, plus the processor and memory space required to execute the query platform 10, is still lower than the processor load and memory space of the electronic devices 20 and 30, then the Figure 3 architecture is adopted.

[0077] Please also note that the server 35 here is only used as an example. In actual application, one or more servers 35 for implementing the function of the query platform 10 may be used, and the locations of the servers 35 may be the same or different.

[0078] Alternatively, when the manufacturer determines that, under normal circumstances of the user operating other software, the sum of the space of the processor and memory required, plus the space of the processor and memory required to execute the language translator 11, is still lower than the number of processors and the memory space actually provided by the electronic device. However, if the semantic core module 13 and / or the pre-trained transformer 15 are further added together, and the processor load and memory space occupied by the query function exceed the number of processors and the memory space actually provided by the electronic device 30, the language translator 11 is set on the electronic device 30, and the semantic core module 13 and the pre-trained transformer 15 are set on the cloud. The implementation method of which parts of the query platform 10 should be set on the electronic device 30 and which parts should be set on the server 35 can be quite flexible and variable, and will not be elaborated in detail in this article.

[0079] Please refer to Figure 4 , which is a relational diagram of processing and converting data by the query platform conceived according to the present invention. This attached drawing is based on Figure 1 's architecture and further illustrates the operations related to the language translator 11 and the semantic core module 13. These components in this attached drawing can be signal-connected to each other, electrically connected to each other, or part of them are signal-connected and the other part is electrically connected according to different applications.

[0080] As mentioned above, the multiple pre-stored token vectors TOKvec stored in the vector database 131 have been pre-established and stored in the vector database 131 by the syntax analyzer 18c and the word vector model 18g before the electronic devices 20 and 30 leave the factory. Therefore, when the user uses the electronic devices 20 and 30, no matter which architecture among Figure 2 , 3 the query platform 10 adopts, the user can directly use the query platform 10 through the electronic devices 20 and 30 without additionally matching the syntax analyzer 18c and the word vector model 18g.

[0081] In addition to the vector database 131 and the prompt information adjustment module 133, according to the semantic core module 13 conceived by the present invention, a vector data cache module 131a can be further set. Briefly speaking, the vector data cache module 131a can be regarded as a buffer storage space. That is, the highly relevant token vector TOKvec_rel generated by the vector database 131 that a certain user has previously asked is stored in a specific or dedicated storage space.

[0082] In this way, if the user wants to query the operation content in the future and has previously queried similar or relevant operations, the highly relevant token vector TOKvec_rel that has previously appeared and been stored can be found in the vector data cache module 131a, without having to repeat the query in the vector database 131. For example, when the user previously queried the screen resolution setting, the vector data cache module 131a may record the highly relevant token vector TOKvec_rel related to the term "display settings". Later, if the user wants to perform a dual-screen setting, the highly relevant token vector TOKvec_rel related to the term "display settings" can be provided by the vector data cache module 131a. There is no need to limit the substitution method for optimizing or accelerating the query process of the vector database 131 in this part.

[0083] From Figure 4 It can be seen that the language translator 11 includes: a language identifier 111 and a language translation model 113. Among them, the language translation model 113 further includes: an input translation model (intrMDL) 113a and an output translation model (otrMDL) 113c.

[0084] After the user generates a non-English query string qryTXT(usrLNG) using the input device, the non-English query string qryTXT(usrLNG) is transmitted to the language identifier 111 and the input translation model (intrMDL) 113a. The language identifier 111 is used to identify and determine the user's preferred language usrLNG corresponding to the non-English query string qryTXT(usrLNG), and accordingly generate a language identifier usrLNG_ID. The language identifier 111 transmits the language identifier usrLNG_ID to the input translation model (intrMDL) 113a and the output translation model (otrMDL) 113c.

[0085] The input translation model (intrMDL) 113a receives the non-English query string qryTXT(usrLNG) from the input device and the language identifier usrLNG_ID from the language identifier 111. According to the language identifier usrLNG_ID, the input translation model (intrMDL) 113a translates the non-English query string qryTXT(usrLNG) into an English query string qryTXT(eng).

[0086] The input translation model (intrMDL) 113a transmits the English query string qryTXT(eng) to the vector data buffer module 131a and / or the vector database 131. If the vector data buffer module 131a stores the highly relevant token vector TOKvec_rel related to the English query string qryTXT(eng), the highly relevant token vector TOKvec_rel is transmitted by the vector data buffer module 131a. Conversely, if the vector data buffer module 131a does not store the highly relevant token vector TOKvec_rel related to the English query string qryTXT(eng), the input translation model (intrMDL) 113a and / or the vector data buffer module 131a can transmit the English query string qryTXT(eng) to the vector database 131, and the vector database 131 generates the highly relevant token vector TOKvec_rel according to the English query string qryTXT(eng).

[0087] Thereafter, the prompt message adjustment module 133 converts the English query string qryTXT(eng) into an English interactive prompt string pptTXT(eng) that is easier for the pre-trained transformer 15 to understand based on the highly relevant token vector TOKvec_rel. Moreover, the pre-trained transformer 15 infers and generates an English response string rTXT(eng) in response to the English interactive prompt string pptTXT(eng), and transmits the English response string rTXT(eng) to the output translation model (otrMDL) 113c.

[0088] The output translation model (otrMDL) 113c receives the English response string rTXT(eng) from the pre-trained transformer 15 and the language identifier usrLNG_ID from the language identifier 111. According to the language identifier usrLNG_ID, the output translation model (otrMDL) 113c translates the English response string rTXT(eng) into a non-English response string rTXT(usrLNG).

[0089] In actual application, the language identifier 111 can also be optional. That is, the electronic device provides a setting option for the preferred language, allowing the user to pre-select their preferred user preferred language usrLNG. Alternatively, the language system used by the operating systems of the electronic devices 20 and 30 can be directly used as the user preferred language usrLNG. In such cases, the language identifier usrLNG_ID is a default value and does not need to be determined by the language identifier 111. Therefore, in Figure 4In the figure, the language identifier 111 is drawn in dashed lines, along with the non-English query string qryTXT(usrLNG) input to the language identifier 111 and the language identifier usrLNG_ID output by the language identifier 111. Details of this application-related change are not elaborated in this article.

[0090] Incidentally, if the user's preferred language usrLNG is English, the language translator 11 of the query platform 10 may not be used. For example, the language translator 11 can be directly bypassed, and the information input by the user from the input device can be directly regarded as the English query string qryTXT(eng); and the English response string rTXT(eng) can be directly played using the output device. This part belongs to application-related changes and does not need to be restricted.

[0091] Before the electronic devices 20 and 30 leave the factory, the manufacturers of the electronic devices 20 and 30 pre-convert the content of the English version user manual 18a into the format of the pre-stored token vector TOKvec using the parser 18c and the word vector model 18g. Subsequently, these pre-stored token vectors TOKvec are stored in the vector database 131. Depending on the specifications of the electronic devices 20 and 30 themselves, the vector database 131 can be built into the electronic device 20 (local), or set on the server 35. After the electronic devices 20 and 30 leave the factory, if the user wishes to query the operation instructions, the electronic devices 20 and 30 will use Figure 4 the query platform 10 to execute Figure 5A 、 5B the process.

[0092] Please refer to Figure 5A 、 5B which is a flowchart of the query method conceived according to the present invention. Please also refer to Figure 4 、 5A 、5B. First, when the electronic devices 20 and 30 are just powered on or the query platform 10 is just enabled, the electronic devices 20 and 30 need to perform an initial setup on the query platform 10 (step S301). The process of initial setup refers to the process of establishing the usage environment of the query platform 10 after the electronic devices 20 and 30 are powered on until the user can start using the query function. For example, the process of initial setup may include: setting the language translator 11, loading the language translator 11 into the semantic core module 13, starting and loading the vector database 131, the process of establishing a connection with the pre-trained transformer 15, etc.

[0093] Next, the query platform 10 determines whether the user wants to query the operation instructions (step S303). If not, after waiting for a period of time, step S303 is executed again. Conversely, if the user wishes to query the operation instructions related to the electronic devices 20 and 30, the query platform 10 receives the non-English query string qryTXT(usrLNG) input by the user through the input device (step S305).

[0094] After the language identifier 111 and the input translation model (intrMDL) 113a receive the non-English query string qryTXT(usrLNG) (step S307), the language identifier 111 analyzes the user's habitual language usrLNG corresponding to the non-English query string qryTXT(usrLNG). Moreover, the language identifier 111 transmits the language identifier usrLNG_ID representing the user's habitual language usrLNG to the input translation model (intrMDL) and the output translation model (otrMDL) (step S309).

[0095] As mentioned above, in some applications, the user's habitual language usrLNG can be preset on the electronic devices 20 and 30 and does not need to be generated by the language identifier 111. Therefore, the actions of the language identifier 111 mentioned in steps S307 and S309 are optional.

[0096] Based on the language identifier usrLNG_ID, the input translation model (intrMDL) 113a translates the non-English query string qryTXT(usrLNG) into an English query string qryTXT(eng) (step S311). The English query string qryTXT(eng) is transmitted to the vector database 131 and the prompt information adjustment module 133 (steps S313a, S315).

[0097] After receiving the English query string qryTXT(eng), the vector database 131 retrieves multiple pre-stored token vectors TOKvec stored in the vector database 131 according to the English query string qryTXT(eng). After the vector database 131 retrieves the pre-stored token vectors TOKvec, it generates a highly relevant token vector TOKvec_rel (step S313c).

[0098] The hint message adjustment module 133 converts the English query string qryTXT(eng) into an English interactive hint pptTXT(eng) according to the highly relevant token vector TOKvec_rel provided by the vector database 131 (step S317). The pre-trained transformer 15 generates an English reply string rTXT(eng) for the content of the English interactive hint pptTXT(eng) (step S319). The output translation model (otrMDL) 113c translates the English reply string rTXT(eng) into a non-English reply string rTXT(usrLNG) corresponding to the user's preferred language usrLNG according to the language identifier usrLNG_ID (step S321). The electronic devices 20, 30 can then provide the content of the non-English reply string rTXT(usrLNG) to the user through voice playback, screen display, or other means. Thereafter, the query platform 10 determines again whether the user still has other operation instructions to query (step S303).

[0099] In summary, the query system 1 of the present invention can generate a non-English reply string rTXT(usrLNG) also displayed in the user's preferred language for a non-English query string qryTXT(usrLNG) input by the user in their preferred language. Accordingly, the user can maintain the use of a familiar operation interface. On the other hand, the content of the non-English reply string rTXT(usrLNG) generated by the query system 1 has been pre-referred based on the content of the English version of the user manuals 18a of the electronic devices 20, 30 operated by the user. Therefore, in terms of the actual query results on how to operate, the non-English reply string rTXT(usrLNG) heard or seen by the user is directly related to the electronic devices 20, 30 they operate and is a specific operation instruction.

[0100] The query method according to the concept of the present invention can be executed by a software program stored in a computer program product or a computer-readable medium. For users who are not familiar with the operation of the electronic devices 20, 30, when querying operation instructions, they can not only use their preferred language, but also because the content replied by the query platform 10 also originates from the content of the English version of the user manuals 18a of the electronic devices 20, 30, the operation suggestions replied by the query platform 10 will not be empty or inaccurate. On the other hand, for the manufacturers of the electronic devices 20, 30, there is no need to prepare user manuals in different language versions for users in different countries, which can greatly reduce production costs.

[0101] In actual application, the query method for the operation instructions of the present invention can be applied to electronic devices such as mobile phones, tablets, desktop computers, and laptop computers. For the electronic devices that can be paired with the query method for the operation instructions of the present invention, they can all be freely substituted by the known technologies in the technical field to which this case belongs.

[0102] Those with ordinary knowledge in the art can all understand that in the above description, various logic blocks, modules, circuits, and method steps used as examples can all be implemented by using electronic hardware, computer software, or a combination of the two. Moreover, for the connection methods among these implementation methods, regardless of whether the terms such as signal link, connection, coupling, electrical connection, or other types of alternative practices are used in the above description, their purpose is only to illustrate that when implementing logic blocks, modules, circuits, and method steps, different means can be used, such as wired electronic signals, wireless electromagnetic signals, and optical signals, etc., to directly or indirectly perform signal exchange, and thus achieve the purpose of signal, data, and control information exchange and transmission. Therefore, the terms used in the specification will not form a limitation on the implementation of the connection relationship in this case, nor will it deviate from the scope of this case due to different connection methods.

[0103] In summary, although the present invention has been disclosed above with embodiments, it is not intended to limit the present invention. Those with ordinary knowledge in the technical field to which the present invention belongs can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to that defined by the appended claims.

Claims

1. A query method for operation instructions, comprising the following steps: A language translator translates a non-English query string into an English query string according to a language identifier; A prompt information adjustment module converts the English query string into an English interactive prompt string according to at least one highly relevant token vector, where the at least one highly relevant token vector is related to the operation instructions of an electronic device; and The language translator translates an English reply string inferred based on the English interactive prompt string into a non-English reply string according to the language identifier.

2. The query method according to claim 1, wherein The English reply string is generated by a pre-trained transformer inferring on the English interactive prompt string.

3. The query method according to claim 1, wherein It further comprises the following steps: A parser decomposes sentences and segments tokens in at least one English document file recording the operation instructions of the electronic device to generate at least one JavaScript Object Notation (JSON) file; A word vector model performs word vectorization processing on the at least one JSON file to generate a plurality of pre-stored token vectors; and A vector database stores these pre-stored token vectors.

4. The query method according to claim 3, wherein It further comprises the following steps: The vector database retrieves these pre-stored token vectors according to the English query string to obtain the at least one highly relevant token vector; The prompt information adjustment module receives the English query string from the language translator; and The prompt information adjustment module receives the at least one highly relevant token vector from the vector database.

5. The query method according to claim 1, wherein It further comprises the following steps: The language translator identifies the language of the non-English query string to generate the language identifier.

6. The query method according to claim 1, wherein The electronic device includes an input device, and the input device generates the non-English query string transmitted to the language translator in response to a user's operation.

7. The query method according to claim 6, wherein The language identifier corresponds to a user's preferred language.

8. The query method according to claim 1, wherein The electronic device includes an output device, and the output device plays the non-English reply string received from the language translator.

9. A computer program product, on which a software program is stored. When the software program is executed, it performs a query method for operation instructions, where the query method includes the following steps: Translate a non-English query string into an English query string according to a language identifier; Convert the English query string into an English interactive prompt string according to at least one highly relevant token vector, where the at least one highly relevant token vector is related to the operation instructions of an electronic device; and Translate an English reply string inferred based on the English interactive prompt string into a non-English reply string according to the language identifier.

10. A query system, comprising: A query platform, comprising: A language translator, comprising: An input translation model, which translates a non-English query string into an English query string according to a language identifier; and An output translation model; and A semantic core module that converts the English query string into an English interactive prompt string according to at least one highly relevant token vector. Wherein the at least one highly relevant token vector is related to the operation instructions of an electronic device, and the output translation model translates an English response string inferred based on the English interactive prompt string into a non-English response string according to the language identifier.

Images