Information query method and device, electronic equipment and medium

Abstract

The invention relates to an information query method and device, electronic equipment and a medium. The information query method comprises the following steps: in response to a motion-related query instruction input by a user, inputting the query instruction into a first preset model to obtain vectorized first information; determining vectorized second information matched with the first information; the first information and the second information are input into a second preset model to obtain a first query result, and the first query result is used for guiding the user to move; and sending the first query result to the user. The query instruction is processed through the first preset model and the second preset model to obtain the first query result, and the first query result can guide the movement of the user, so that the effectiveness of information query is improved. Meanwhile, the first information and the second information are vectorized information, so that the information of the user is prevented from being leaked in the process of generating the first query result, and the security of information query is improved.

Description

Technical Field

[0001] This disclosure relates to the field of information retrieval technology, and in particular to an information retrieval method, apparatus, electronic device, and medium. Background Technology

[0002] With the development of electronic devices, they can collect users' motion information and issue corresponding motion information in response to user queries. However, since electronic devices can only issue motion information as a query result and cannot provide guidance for users' movements, the effectiveness of information queries is low. Summary of the Invention

[0003] To overcome the problems existing in related technologies, this disclosure provides an information query method, apparatus, electronic device, and medium.

[0004] According to a first aspect of the present disclosure, an information query method is provided, the information query method comprising:

[0005] In response to a user's input of a motion-related query command, the query command is input into a first preset model to obtain vectorized first information;

[0006] Determine the vectorized second information that matches the first information;

[0007] The first information and the second information are input into the second preset model to obtain the first query result, which is used to guide the user's exercise.

[0008] Send the first query result to the user.

[0009] In some embodiments of this disclosure, the first preset model includes a large sensor model and an alignment model; the step of inputting the query command into the first preset model to obtain vectorized first information includes:

[0010] The query command is input into the sensor large model to obtain vectorized third information;

[0011] The third information is input into the alignment model to obtain the first information in natural language vectorization.

[0012] In some embodiments of this disclosure, the first preset model further includes multiple linear detection layers, each linear detection layer corresponding to a sensor; the step of inputting the query command into the sensor large model to obtain vectorized third information includes:

[0013] When the type of the query instruction is the first type, the query instruction is input into the sensor large model to obtain the third information, where the first type is a type related to motion guidance;

[0014] When the type of the query instruction is the second type, the first motion information collected by the corresponding sensor is determined according to the query instruction, where the second type is a type related to information collection;

[0015] The first motion information is input into the sensor large model and passed through the corresponding linear detection layer to obtain the third information.

[0016] In some embodiments of this disclosure, the step of inputting the first information and the second information into a second preset model to obtain a first query result includes:

[0017] When the type of the query instruction is the first type, the first information and the second information are input into the second preset model to obtain the first query result;

[0018] The information query method also includes:

[0019] When the type of the query instruction is the second type, the first information is input into the second preset model to obtain the second query result;

[0020] The second query result is sent to the user.

[0021] In some embodiments of this disclosure, the information query method further includes:

[0022] Acquire fourth information from multiple sensors in different electronic devices;

[0023] The discrete fourth pieces of information are converted into a series of consecutive fifth pieces of information;

[0024] The sensor large model is obtained by using the fifth information described above for model training.

[0025] In some embodiments of this disclosure, converting the discrete fourth pieces of information into a plurality of consecutive fifth pieces of information includes:

[0026] The fourth information collected by each sensor and the corresponding label information of the sensor are mapped as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information;

[0027] Each set of the sixth information is serialized to obtain multiple sets of seventh information;

[0028] After integrating the seventh information from each group, dimensionality reduction processing is performed to obtain the fifth information from each group.

[0029] In some embodiments of this disclosure, the serialization process performed on each group of the sixth information to obtain multiple groups of seventh information includes:

[0030] Different objective functions are used to positionally encode the sixth information of the corresponding group to obtain the seventh information of each group;

[0031] Wherein, the frequency parameter of the objective function is the acquisition frequency of the sensor corresponding to the sixth information.

[0032] In some embodiments of this disclosure, the step of using the fifth information to train the model and obtain the sensor large model includes:

[0033] The sensor large model is obtained by using a portion of the fifth information as the training set and another portion of the fifth information as the validation set for model training.

[0034] In some embodiments of this disclosure, after the model is trained using the fifth information to obtain the large sensor model, the information query method further includes:

[0035] Multiple linear detection layers are constructed, each linear detection layer corresponding to a sensor;

[0036] The corresponding linear detection layer is trained using the fourth information and the sensor large model.

[0037] In some embodiments of this disclosure, after the model is trained using the fifth information to obtain the large sensor model, the information query method further includes:

[0038] Each of the fourth pieces of information is input into the sensor large model to obtain the vectorized eighth pieces of information;

[0039] The alignment model is obtained by training the model using the eighth piece of information and the natural language information corresponding to the eighth piece of information.

[0040] In some embodiments of this disclosure, the second preset model is a large language model; the step of inputting the first information and the second information into the second preset model to obtain the first query result includes:

[0041] Based on the first information and the second information, construct a prompt template;

[0042] The prompt template is input into the large language model to obtain the first query result described in natural language text.

[0043] In some embodiments of this disclosure, the information query method further includes:

[0044] Collect the user's second motion information;

[0045] The second motion information is input into the first preset model to obtain at least a portion of the information in the vector database; and / or,

[0046] To acquire knowledge and information related to sports;

[0047] Inputting the knowledge information into the first preset model yields at least a portion of the information in the vector database; and / or,

[0048] Input the historical query results into the first preset model to obtain at least some information from the vector database;

[0049] The second information is information from the vector database.

[0050] According to a second aspect of the present disclosure, an information query device is provided, the information query device comprising:

[0051] The first processing module is configured to respond to a motion-related query command input by a user, input the query command into a first preset model, and obtain vectorized first information.

[0052] A determining module, configured to determine vectorized second information that matches the first information;

[0053] The second processing module is configured to input the first information and the second information into a second preset model to obtain a first query result.

[0054] The issuing module is configured to issue the first query result to the user.

[0055] According to a third aspect of the present disclosure, an electronic device is provided, the electronic device comprising:

[0056] processor;

[0057] Memory used to store the processor's executable instructions;

[0058] The processor is configured to execute the information query method described above.

[0059] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, which, when instructions in the storage medium are executed by a processor of a terminal, enables the terminal to perform the information query method as described above.

[0060] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects:

[0061] In response to a user's input of a motion-related query, the user needs to retrieve motion-related information. The query is input into a first preset model, resulting in vectorized first information to reduce the complexity of the query and enable a second preset model to recognize it. Vectorized second information matching the first information is determined to obtain motion-related information. The first and second information are then input into the second preset model, which generates a motion-related first query result based on the first and second information. This first query result is then sent to the user to inform them of the motion-related information. By processing the query through the first and second preset models to obtain the first query result, the first query result can guide the user's movement, thereby improving the effectiveness of information retrieval. Simultaneously, because the first and second information are vectorized, the generation of the first query result avoids disclosing user information, thus improving the security of information retrieval.

[0062] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0063] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0064] Figure 1 This is a flowchart illustrating an information query method according to an exemplary embodiment;

[0065] Figure 2 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0066] Figure 3 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0067] Figure 4 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0068] Figure 5 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0069] Figure 6 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0070] Figure 7 This is a flowchart illustrating an information query method according to another exemplary embodiment;

[0071] Figure 8This is a block diagram illustrating an information query device according to an exemplary embodiment;

[0072] Figure 9 This is a block diagram of an electronic device according to an exemplary embodiment.

[0073] In the picture:

[0074] 100 - First processing module; 150 - Determining module; 200 - Second processing module; 250 - Issuing module; 400 - Electronic device; 402 - Processing component; 404 - Memory; 406 - Power supply component; 408 - Multimedia component; 410 - Audio component; 412 - Input / output interface; 414 - Sensor component; 416 - Communication component; 420 - Processor. Detailed Implementation

[0075] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims. It should also be understood that the term “and / or” as used in this disclosure refers to and includes any or all possible combinations of one or more of the associated listed items.

[0076] As users become increasingly health-conscious, more and more are exercising during their leisure time to improve their physical fitness. Electronic devices can collect and store exercise information during activity, and can then provide corresponding exercise information based on user queries. For example, a heart rate sensor can collect and display heart rate data during exercise. However, because electronic devices can only provide the exercise information collected by the sensors as a query result, they cannot combine this information with the user's specific needs to provide guidance on their exercise, resulting in low effectiveness of the information query. For instance, if a user queries how to exercise to reduce body fat percentage, the electronic device cannot provide a query result that incorporates the exercise information collected by the sensors.

[0077] To address the aforementioned technical issues, this disclosure provides an information query method. A first query result is generated by combining first information transformed from a motion-related query instruction with second information matching the first information. This first query result informs the user how to perform the exercise to achieve their goal, thereby improving the effectiveness of the information query. Furthermore, since the first and second information are vectorized, information leakage from the user is avoided, thus enhancing the security of the information query.

[0078] This disclosure provides an information query method, such as... Figure 1 As shown, the method includes:

[0079] S100: In response to the user's input of a motion-related query command, the query command is input into the first preset model to obtain vectorized first information.

[0080] S200, Determine the vectorized second information that matches the first information.

[0081] S300. Input the first information and the second information into the second preset model to obtain the first query result. The first query result is used to guide the user's movement.

[0082] S400: Send the first query result to the user.

[0083] In this embodiment, in response to a user's input of a motion-related query command, the user needs to query motion-related information. The query command is input into a first preset model to obtain vectorized first information, reducing the complexity of the query command and enabling a second preset model to recognize it. Vectorized second information matching the first information is determined to obtain motion-related information for the user. The first and second information are input into the second preset model, which generates a motion-related first query result based on the first and second information. The first query result is then sent to the user to inform them of the motion-related information. By processing the query command through the first and second preset models to obtain the first query result, the first query result can guide the user's movement, thereby improving the effectiveness of information retrieval. Simultaneously, because the first and second information are vectorized, the generation of the first query result avoids disclosing user information, thus improving the security of information retrieval.

[0084] In one embodiment, such as Figure 2 As shown, the first preset model includes a large sensor model (LSM) and an alignment model. The vectorized first information obtained by inputting the query command into the first preset model in step S100 is determined as follows:

[0085] S110. Input the query command into the sensor large model to obtain vectorized third information.

[0086] S120. Input the third information into the alignment model to obtain the first information of natural language vectorization.

[0087] In this embodiment, the query command is input into a large sensor model, which then vectorizes the query command to reduce its complexity, resulting in vectorized third information. This third information is then input into an alignment model to align it into information recognizable by a second preset model, yielding vectorized first information in natural language. By processing the query command using both the large sensor model and the alignment model to obtain the first information, which has low complexity and can be recognized by the second preset model, the reliability of information retrieval is improved.

[0088] For example, the alignment model can be a fusion visual-semantic model (Querying Transformer, Q-Former).

[0089] In one embodiment, the first preset model further includes multiple linear probing layers, each corresponding to a sensor. The vectorized third information obtained by inputting the query command into the sensor large model in step S110 is determined as follows:

[0090] If the query command type is type 1, input the query command into the sensor large model to obtain the third information. Type 1 is the type related to motion guidance.

[0091] When the query instruction is of type two, the first motion information collected by the corresponding sensor is determined according to the query instruction. Type two is a type related to information collection.

[0092] The first motion information is input into the sensor's large model and passed through the corresponding linear detection layer to obtain the third information.

[0093] In this embodiment, when the query instruction is of type one, the user needs motion guidance and inputs the query instruction into the sensor large model to obtain third information. When the query instruction is of type two, the user needs to query the first motion information collected by the sensor and determine the first motion information corresponding to the query instruction. The first motion information is input into the sensor large model and passed through the corresponding linear detection layer. The sensor large model and the linear detection layer correct the first motion information to obtain the third information. By processing the query instruction in a corresponding manner for different types of query instructions, the first query result is avoided to be irrelevant to the query instruction, thereby improving the reliability of information query. At the same time, processing the second type of query instruction through the sensor large model and the corresponding linear detection layer can correct the first motion information collected by the sensor, thereby improving the accuracy of information query.

[0094] For example, the first type of query instruction can be a query for exercise-related information other than exercise information. That is, an instruction to obtain exercise guidance. For example, the first type of query instruction can be how much longer I need to exercise today to maintain good health, or what kind of exercise I need to do today to lose weight within a month. The second type of query instruction can be a query for exercise information collected by sensors. For example, the second type of query instruction can be a query for today's average heart rate, or a query for today's step count.

[0095] For example, the step of inputting the first motion information into the sensor large model and passing it through the corresponding linear detection layer to obtain the third information can be replaced by inputting the first motion information and the tag information of the sensor corresponding to the first motion information into the sensor large model and passing it through the corresponding linear detection layer to obtain the third information. The sensor tag information is used to enable the sensor large model to identify the sensor that needs calibration and select the corresponding linear detection layer.

[0096] In one embodiment, the first query result obtained by inputting the first information and the second information into the second preset model in step S300 can be determined in the following way:

[0097] If the query instruction type is the first type, the first information and the second information are input into the second preset model to obtain the first query result.

[0098] Information retrieval methods also include:

[0099] If the query instruction type is the second type, the first information is input into the second preset model to obtain the second query result.

[0100] Send the second query result to the user.

[0101] In this embodiment, when the query instruction type is the first type, the user needs motion guidance and inputs the first information and the matched second information into the second preset model to obtain the first query result. When the query instruction type is the second type, the user needs to query the first motion information collected by the sensor and inputs the corrected first information into the second preset model to obtain the second query result. By obtaining query results in corresponding ways for different query instruction types, the complexity of motion information query can be reduced by obtaining motion guidance through matching or without matching, thereby improving the reliability of information query.

[0102] For example, sending the first query result to the user in step S400 and sending the second query result to the user in the above steps may include displaying the first query result or the second query result, playing voice information of the first query result or the second query result, etc.

[0103] In one embodiment, such as Figure 3 As shown, the second preset model is a Large Language Model (LLM). The step S300 of inputting the first and second information into the second preset model to obtain the first query result can also be determined in the following way:

[0104] S310. Based on the first information and the second information, construct a prompt template.

[0105] S320. Input the prompt template into the large language model to obtain the first query result described in natural language text.

[0106] In this embodiment, since the large language model needs to generate corresponding results based on the prompt template, a prompt template is constructed based on the first and second information. The prompt template is input into the large language model to obtain the first query result described in natural language text, thus informing the user of the first query result through natural language description. Obtaining the first query result through the large language model facilitates user understanding of the first query result, thereby improving the reliability of information retrieval.

[0107] For example, the second information can be information from a vector database, such as second motion information processed by the first preset model, motion-related knowledge information processed by the first preset model, or historical query results. Determining the vectorized second information matching the first information in step S200 can be done by querying the vector database using methods such as brute-force query (IndexFlatL2), spatial clustering (IndexVFFlat), storage optimization (IndexVFPQ), or graph indexing (IndexHNSWFlat). The vectorized second information matching the first information can be multiple pieces of information related to the first information in the vector database. For example, if the first information is "What kind of exercise should I do today to lose weight within a month?", the second information could include the user's heart rate information, height information, weight information, commonly used exercise methods, calorie consumption for each exercise method, and the duration of each exercise session.

[0108] For example, the first motion information and the second motion information may include information collected by one or more sensors when the user is moving, information collected by one or more sensors when the user is sitting still, and physiological information input by the user, etc.

[0109] For example, the large sensor model and alignment model can be set on an electronic device. The large language model can be set on a cloud server or on an electronic device. Since the first and second information are vectorized, even if they are uploaded to a cloud server for conversion, the user's information will not be leaked, thereby improving the security of information retrieval.

[0110] In one embodiment, the information query method further includes:

[0111] Collect the user's second motion information;

[0112] The second motion information is input into the first preset model to obtain at least some information from the vector database.

[0113] The second piece of information is from the vector database.

[0114] In this embodiment, since the first query result relies on information in the vector database, the vector database needs to be pre-established before querying motion-related information. During the user's use of the electronic device, second motion information is collected by multiple sensors to obtain the user's daily motion information as a reference. The second motion information is input into a first preset model to obtain at least a portion of the information in the vector database, which is then matched to determine the second information in a format consistent with the first information. By converting the second motion information into at least a portion of the information in the vector database, the first query result uses the user's motion information as a reference, thereby improving the accuracy of the first query result.

[0115] In one embodiment, the information query method further includes:

[0116] Obtain knowledge and information related to sports.

[0117] By inputting knowledge information into the first preset model, at least some information from the vector database is obtained.

[0118] The second piece of information is from the vector database.

[0119] In this embodiment, since the first query result relies on information in the vector database, the vector database needs to be pre-established before querying motion-related information. Because the first query result depends on motion-related knowledge information, the relationship between the query instruction and the first query result is established through this knowledge information to obtain motion-related knowledge information. This knowledge information is input into a first preset model to obtain at least a portion of the information in the vector database, which is then matched to determine the second information in a form consistent with the first information. By converting the knowledge information into at least a portion of the information in the vector database, the first query result is based on this knowledge information, thereby improving the reliability of the first query result.

[0120] For example, the knowledge information may include the calorie consumption corresponding to each type of exercise, the relationship between calories and body weight, the calories of each food, and the relationship between calories and body weight.

[0121] In one embodiment, the information query method further includes:

[0122] Input the historical query results into the first preset model to obtain at least some information from the vector database.

[0123] The second piece of information is from the vector database.

[0124] In this embodiment, since the first query result relies on information in the vector database, the vector database needs to be pre-established before querying sports-related information. Because historical query results can reflect a user's exercise habits and physical fitness, these results are input into a first preset model to obtain at least a portion of the information in the vector database. By converting historical query results into at least a portion of the information in the vector database, the first query result matches the user's exercise habits and physical fitness, thereby improving the accuracy of the first query result.

[0125] For example, historical query results may include previously obtained first query results and / or second query results.

[0126] In one embodiment, such as Figure 4 As shown, information retrieval methods also include:

[0127] S500: Acquire fourth information collected by multiple sensors in different electronic devices.

[0128] S510. Convert the discrete fourth information into multiple consecutive fifth information.

[0129] S520: Use the fifth information to train the model and obtain the large sensor model.

[0130] In this embodiment, before processing the query command using the large sensor model, the model needs to be trained in advance to obtain the large sensor model. Since the types, numbers, and locations of sensors differ in different electronic devices, to ensure the universality of the large sensor model, fourth information collected by multiple sensors in different electronic devices is acquired and analyzed. Because the fourth information collected by each sensor is discrete and difficult to use for model training, the discrete fourth information is converted into multiple continuous fifth information. The model is trained using each fifth information, enabling the large sensor model to recognize the input information for prediction and processing, thus obtaining the large sensor model. By obtaining the large sensor model based on the fourth information collected by each sensor in different electronic devices, the large sensor model is applicable to various electronic devices and is unaffected by different sensors, thereby improving the accuracy of the large sensor model. Furthermore, using the large sensor model eliminates the need to train a separate large model for each electronic device, thus improving the universality of the large sensor model.

[0131] For example, sensors may include accelerometers, gyroscopes, magnetometers, barometers, distance sensors, light sensors, sound sensors, GPS sensors, heart rate sensors, etc. That is, sensors may include, but are not limited to, sensors used to collect motion information. During the training of the large sensor model, the sensors can collect fourth information at their respective highest acquisition frequencies.

[0132] In one embodiment, such as Figure 5 As shown, the process of converting the discrete fourth pieces of information into multiple consecutive fifth pieces of information in step S510 is determined in the following way:

[0133] S511. Map the fourth information collected by each sensor and the corresponding sensor tag information as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information.

[0134] S512. Serialize each group of sixth information to obtain multiple groups of seventh information.

[0135] S513. After integrating the seventh information of each group, perform dimensionality reduction processing to obtain the fifth information of each group.

[0136] In this embodiment, the fourth information collected by each sensor and the corresponding sensor label information are treated as a set of information to associate the fourth information with the sensor during processing. Since the fourth information is low-dimensional and discrete, making it difficult to use for model training, each set of information is mapped to a high-dimensional vector space, resulting in multiple sets of high-dimensional, dense sixth information. Because each set of fourth information is collected at different times, each set of sixth information is serialized to obtain multiple sets of seventh information reflecting the order of collection. The sets of seventh information are then integrated and subjected to dimensionality reduction to avoid excessively high dimensionality, resulting in individual sets of fifth information. By processing the fourth information into continuous and low-dimensional fifth information, the complexity of the fifth information is reduced, thereby improving the reliability of the large sensor model. Simultaneously, the low dimensionality and complexity of the fifth information allow the trained large sensor model to be deployed on electronic devices, thus improving the security of information retrieval.

[0137] For example, in step S511, the fourth information collected by each sensor and the corresponding sensor's label information are mapped as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information. Each sensor can correspond to an embedding layer, and the information mapping is performed through the embedding layer. For instance, if the accelerometer collects information 512 times, each time in three dimensions, there are a total of 3*512 pieces of fourth information. The accelerometer's label information is 1, and the 513th three-dimensional information is also 1, resulting in a set of 3*513 pieces of information. The embedding layer maps these 3*513 pieces of information to a high-dimensional vector space, resulting in 3*513*768 vector pieces of information as a set of sixth information. That is, each piece of information in each dimension is mapped to 768 vector pieces of information. The label information differs for different sensors. It is understood that the accelerometer's label information 1, the number of collections (512), and the values ​​in the 768 vector pieces of information can be adjusted as needed; the above is merely an example. If an electronic device does not have a certain sensor (such as not including a distance sensor), then the corresponding fourth information and tag information are both set to 0.

[0138] In one embodiment, the serialization process performed on each group of sixth information in step S512 to obtain multiple groups of seventh information is determined in the following manner:

[0139] Different objective functions are used to positionally encode the sixth information of the corresponding group to obtain the seventh information of each group.

[0140] The frequency parameter of the objective function is the sensor acquisition frequency corresponding to the sixth piece of information.

[0141] In this embodiment, since each sensor may acquire the fourth information at different frequencies, and these frequencies are related to the sensor's power consumption, the large sensor model needs to adapt to the acquisition frequencies of different sensors. During the serialization process of each sixth piece of information, objective functions with different frequency parameters are used to positionally encode the sixth information of corresponding groups to adapt to the acquisition frequencies, resulting in the seventh information for each group. By combining the sensor's acquisition frequency with positional encoding of the sixth information, errors in the sequence of the sixth information are avoided, and the large sensor model can adapt to multiple sensors, thereby improving the reliability of the large sensor model.

[0142] For example, the objective function can be expressed by the following formula:

[0143]

[0144] in, f represents the sensor's sampling frequency, d model This represents the dimension of the corresponding sixth piece of information.

[0145] For example, in step S513, the integration of the seventh information from each group followed by dimensionality reduction to obtain the fifth information can be achieved by first integrating the seventh information from each group, then performing dimensionality reduction on the seventh information, and finally inputting the dimensionality-reduced information into an encoder-only model to obtain the fifth information. For example, the integration of the seventh information in step S513 can involve integrating the seventh information from multiple sensors, such as the accelerometer (3*513*768 pieces of position-encoded seventh information), the gyroscope (3*513*768 pieces of position-encoded seventh information), the light sensor (1*513*768 pieces of position-encoded seventh information), and the barometer (1*513*768 pieces of position-encoded seventh information), to obtain n*513*768 pieces of information. In step S513, the seventh information from each group is integrated and then subjected to dimensionality reduction. For example, a temporal block can be added before the multi-head attention layer of the encoder-only model to reduce the number of pieces of information from n*513*768 to n*256*768. It is understood that the dimension of 256 can be adjusted as needed; the above is merely an example.

[0146] In one embodiment, the large sensor model obtained by training the model using each piece of fifth information in step S520 is determined in the following way:

[0147] The sensor large model is obtained by using part of the fifth information as the training set and another part of the fifth information as the validation set for model training.

[0148] In this embodiment, by splitting the fifth piece of information for model training, it is possible to train on a large-scale unlabeled dataset, thereby improving the versatility of the large sensor model.

[0149] For example, the above steps, where a portion of the fifth information is used as the training set and another portion as the validation set to train the model and obtain a large sensor model, can be trained using the Masked Sensor Modeling method. For instance, the large sensor model can be obtained by using 60% of the fifth information as the training set and 40% as the validation set, with the model predicting the 40% of the fifth information using 60% of the fifth information.

[0150] In one embodiment, such as Figure 6 As shown, after using the fifth information in step S520 to train the model and obtain the large sensor model, the information query method further includes:

[0151] S530: Construct multiple linear detection layers, each corresponding to a sensor.

[0152] S540 uses fourth information and a large sensor model to train the corresponding linear detection layer.

[0153] In this embodiment, since the sensor large model is a general large model, the information collected by the sensors may be inaccurate after being input into the sensor large model. Therefore, a linear probe layer is constructed for each sensor. The corresponding linear probe layer is trained using the fourth information and the sensor large model, so that the information collected by the sensors can be corrected after being input into the sensor large model and passed through the corresponding linear probe layer. By adding a linear probe layer to correct the information collected by the sensors, it is not necessary to set up a model for each sensor, thereby reducing the complexity of the sensor large model. At the same time, when the user's query instruction is of the second type, only the linear probe layer is needed for correction, without having to find matching second information, thereby reducing the complexity of information query.

[0154] In one embodiment, after using the fifth information to train the model in step S520 to obtain the large sensor model, the information query method further includes:

[0155] The fourth information is input into the sensor's large model to obtain the vectorized eighth information.

[0156] The alignment model is obtained by training the model using the eighth piece of information and the natural language information corresponding to the eighth piece of information.

[0157] In this embodiment, since the vectorized information output by the large sensor model after the query command is input cannot be recognized by the large language model, an alignment model needs to be added between the large sensor model and the large language model to convert the information. Each fourth piece of information is input into the large sensor model, which processes the fourth piece of information to obtain vectorized eighth information. The model is then trained using the eighth information and its corresponding natural language information to ensure a correspondence between the vectorized eighth information and the natural language information, thus obtaining the alignment model. This alignment model, trained using the eighth information and natural language information, can convert the output of the large sensor model into natural language, enabling the large language model to recognize it, thereby improving the reliability of information retrieval.

[0158] This disclosure provides an information query method, such as... Figure 7 As shown, the method includes:

[0159] S600: Acquire fourth information collected by multiple sensors in different electronic devices.

[0160] S610. Map the fourth information collected by each sensor and the corresponding sensor tag information as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information.

[0161] S620. Different objective functions are used to positionally encode the sixth information of the corresponding group to obtain the seventh information of each group.

[0162] S630. After integrating the seventh information from each group, perform dimensionality reduction processing to obtain the fifth information from each group.

[0163] S640. Using part of the fifth information as the training set and another part of the fifth information as the validation set, the model is trained to obtain the large sensor model.

[0164] S650, construct multiple linear probe layers.

[0165] S660 uses fourth information and a large sensor model to train the corresponding linear detection layer.

[0166] S670. Input each of the fourth pieces of information into the sensor large model to obtain the vectorized eighth information.

[0167] S680. The model is trained using the eighth information and the natural language information corresponding to the eighth information to obtain the alignment model.

[0168] S690: Collect the user's second motion information and obtain motion-related knowledge information.

[0169] S700: Input the second motion information, knowledge information, and historical query results into the sensor large model and obtain the information in the vector database through the alignment model.

[0170] S710, in response to a motion-related query command input by the user, determines the type of the query command.

[0171] S720. If the query instruction is of type 1, input the query instruction into the sensor large model to obtain vectorized third information.

[0172] S730: Input the third information, which has not been passed through the linear probe layer, into the alignment model to obtain the first information of natural language vectorization.

[0173] S740. Construct a prompt template based on the first information and the second information in the vector database that matches the first information.

[0174] S750. Input the prompt template into the large language model to obtain the first query result described in natural language text.

[0175] S760: Send the first query result to the user.

[0176] S770. If the query instruction is of type 2, determine the first motion information collected by the corresponding sensor according to the query instruction.

[0177] S780: Input the first motion information into the sensor large model and pass it through the corresponding linear detection layer to obtain the vectorized third information.

[0178] S790. Input the third information from the linear probe layer into the alignment model to obtain the first information of natural language vectorization.

[0179] S800: Input the first information into the large language model to obtain the second query result.

[0180] S810: Send the second query result to the user.

[0181] In this embodiment, fourth information collected by multiple sensors in different electronic devices is acquired to train a general sensor large-scale model. The fourth information collected by each sensor and the corresponding sensor label information are mapped to a high-dimensional vector space, transforming discrete information into dense information, resulting in multiple sets of sixth information. Different objective functions are used to positionally encode the sixth information in the corresponding sets, serializing each set of sixth information to obtain each set of seventh information. The seventh information is then integrated and dimensionality-reduced to obtain each set of fifth information. A portion of the fifth information is used as the training set, and another portion as the validation set for model training on an unlabeled large-scale dataset, resulting in a sensor large-scale model. Multiple linear probe layers are constructed, and the corresponding linear probe layers are trained using the fourth information and the sensor large-scale model, enabling the sensor large-scale model to work with the linear probe layers to correct the motion information collected by the sensors. Each set of fourth information is input into the sensor large-scale model, which processes the fourth information to obtain vectorized eighth information. The eighth information and its corresponding natural language information are used for model training, ensuring that the vectorized information processed by the sensor corresponds to the natural language information, resulting in an alignment model. The system collects the user's second motion information and acquires motion-related knowledge information. This second motion information, knowledge information, and historical query results are input into a large-scale sensor model and, through an alignment model, vectorized information is obtained to construct a vector database. Responding to the user's motion-related query command, the type of query command is determined to determine the method for obtaining the query result. If the query command type is type one, the query command is input into the large-scale sensor model to obtain vectorized third information. This third information is input into the alignment model to obtain vectorized first information in natural language, enabling the large-scale language model to recognize it. Based on the first information and the second information matching it, a prompt template is constructed and input into the large-scale language model to obtain and issue a first query result described in natural language text. If the query command type is type two, the corresponding first motion information collected by the sensor is determined based on the query command. This first motion information is input into the large-scale sensor model and, through the corresponding linear detection layer, corrected to obtain vectorized third information. This third information is input into the alignment model to obtain vectorized first information in natural language, enabling the large-scale language model to recognize it. Finally, the first information is input into the large-scale language model to obtain and issue a second query result. The query command is processed using a large sensor model, an alignment model, and a large language model to obtain the first query result. This first query result can guide the user's movement, thereby improving the effectiveness of information retrieval. Simultaneously, because the first and second information are vectorized, the generation of the first query result avoids the leakage of user information, thus improving the security of information retrieval.

[0182] In one exemplary embodiment, an information query device is provided for implementing the above-described method. (Reference) Figure 8 As shown, the information query device may include a first processing module 100, a determining module 150, a second processing module 200, and an issuing module 250. During the implementation of the above method,

[0183] The first processing module 100 is configured to respond to a motion-related query command input by the user, input the query command into a first preset model, and obtain vectorized first information.

[0184] The determination module 150 is configured to determine vectorized second information that matches the first information.

[0185] The second processing module 200 is configured to input the first information and the second information into the second preset model to obtain the first query result.

[0186] The issuing module 250 is configured to issue the first query result to the user.

[0187] In one exemplary embodiment, an information query apparatus is provided, wherein a first processing module 100 is configured to:

[0188] Input the query command into the sensor large model to obtain vectorized third information.

[0189] The third information is input into the alignment model to obtain the first information in the natural language vectorization.

[0190] In one exemplary embodiment, an information query apparatus is provided, wherein a first processing module 100 is configured to:

[0191] If the query command type is type 1, input the query command into the sensor large model to obtain the third information.

[0192] If the query command is of type 2, the first motion information collected by the corresponding sensor is determined according to the query command.

[0193] The first motion information is input into the sensor's large model and passed through the corresponding linear detection layer to obtain the third information.

[0194] In one exemplary embodiment, an information query apparatus is provided, wherein a second processing module 200 is configured to:

[0195] If the query instruction type is the first type, the first information and the second information are input into the second preset model to obtain the first query result.

[0196] In one exemplary embodiment, an information query apparatus is provided, wherein a second processing module 200 is configured to:

[0197] If the query instruction type is the second type, the first information is input into the second preset model to obtain the second query result.

[0198] In one exemplary embodiment, an information query apparatus is provided, wherein an issuing module 250 is configured to:

[0199] Send the second query result to the user.

[0200] In one exemplary embodiment, an information query apparatus is provided, wherein a second processing module 200 is configured to:

[0201] Based on the first and second information, construct a prompt template.

[0202] Input the prompt template into the large language model to obtain the first query result described in natural language text.

[0203] In one exemplary embodiment, an information query device is provided, the device further comprising:

[0204] The acquisition module is configured to acquire the user's second motion information.

[0205] In one exemplary embodiment, an information query apparatus is provided, wherein a first processing module 100 is configured to:

[0206] The second motion information is input into the first preset model to obtain at least some information from the vector database.

[0207] In one exemplary embodiment, an information query device is provided, the device further comprising:

[0208] The acquisition module is configured to acquire knowledge and information related to sports.

[0209] In one exemplary embodiment, an information query apparatus is provided, wherein a first processing module 100 is configured to:

[0210] By inputting knowledge information into the first preset model, at least some information from the vector database is obtained.

[0211] In one exemplary embodiment, an information query apparatus is provided, wherein a first processing module 100 is configured to:

[0212] Input the historical query results into the first preset model to obtain at least some information from the vector database.

[0213] In one exemplary embodiment, an information query apparatus is provided, wherein the acquisition module is configured to:

[0214] Acquire fourth information from multiple sensors in different electronic devices.

[0215] In one exemplary embodiment, an information query device is provided, the device further comprising:

[0216] The training module is configured to convert discrete fourth information into multiple consecutive fifth information.

[0217] The sensor large model is obtained by using the fifth piece of information.

[0218] In one exemplary embodiment, an information query apparatus is provided, wherein a training module is configured to:

[0219] The fourth information collected by each sensor and the corresponding sensor tag information are mapped as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information.

[0220] Each group of sixth information is serialized to obtain multiple groups of seventh information.

[0221] After integrating the seventh information from each group, dimensionality reduction processing is performed to obtain the fifth information from each group.

[0222] In one exemplary embodiment, an information query apparatus is provided, wherein a training module is configured to:

[0223] Different objective functions are used to positionally encode the sixth information of the corresponding group to obtain the seventh information of each group.

[0224] In one exemplary embodiment, an information query apparatus is provided, wherein a training module is configured to:

[0225] The sensor large model is obtained by using part of the fifth information as the training set and another part of the fifth information as the validation set for model training.

[0226] In one exemplary embodiment, an information query apparatus is provided, wherein a training module is configured to:

[0227] Construct multiple linear probe layers.

[0228] The corresponding linear detection layer is trained using fourth information and a large sensor model.

[0229] In one exemplary embodiment, an information query apparatus is provided, wherein a training module is configured to:

[0230] The fourth information is input into the sensor's large model to obtain the vectorized eighth information.

[0231] The alignment model is obtained by training the model using the eighth piece of information and the natural language information corresponding to the eighth piece of information.

[0232] In one exemplary embodiment, an electronic device is provided, such as a mobile phone, a laptop computer, a tablet computer, and a wearable device.

[0233] refer to Figure 9 As shown, the electronic device 400 may include one or more of the following components: processing component 402, memory 404, power supply component 406, multimedia component 408, audio component 410, input / output (I / O) interface 412, sensor component 414, and communication component 416.

[0234] Processing component 402 typically controls the overall operation of electronic device 400, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 402 may include one or more processors 420 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 402 may include one or more modules to facilitate interaction between processing component 402 and other components. For example, processing component 402 may include a multimedia module to facilitate interaction between multimedia component 408 and processing component 402.

[0235] Memory 404 is configured to store various types of data to support the operation of electronic device 400. Examples of this data include instructions for any application or method operating on electronic device 400, contact data, phonebook data, messages, pictures, videos, etc. Memory 404 can be implemented by any type of volatile or non-volatile storage terminal or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0236] Power supply component 406 provides power to various components of electronic device 400. Power supply component 406 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 400.

[0237] Multimedia component 408 includes a screen that provides an output interface between electronic device 400 and user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 408 includes a front-facing camera module and / or a rear-facing camera module. When electronic device 400 is in an operating mode, such as shooting mode or video mode, the front-facing camera module and / or rear-facing camera module may receive external multimedia data. Each front-facing camera module and rear-facing camera module may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0238] Audio component 410 is configured to output and / or input audio signals. For example, audio component 410 includes a microphone (MIC) configured to receive external audio signals when electronic device 400 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 404 or transmitted via communication component 416. In some embodiments, audio component 410 also includes a speaker for outputting audio signals.

[0239] I / O interface 412 provides an interface between processing component 402 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0240] Sensor assembly 414 includes one or more sensors for providing state assessments of various aspects of electronic device 400. For example, sensor assembly 414 may detect the on / off state of electronic device 400, the relative positioning of components such as the display and keypad of electronic device 400, changes in position of electronic device 400 or a component of electronic device 400, the presence or absence of user contact with electronic device 400, orientation or acceleration / deceleration of electronic device 400, and temperature changes of electronic device 400. Sensor assembly 414 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 414 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 414 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0241] Communication component 416 is configured to facilitate wired or wireless communication between electronic device 400 and other terminals. Electronic device 400 can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G, 5G, or combinations thereof. In one exemplary embodiment, communication component 416 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 416 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0242] In an exemplary embodiment, the electronic device 400 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing terminals (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods shown in the above embodiments or combinations thereof.

[0243] In one exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 404 including instructions, which can be executed by a processor 420 of an electronic device 400 to perform the methods shown in the embodiments or combinations thereof. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage terminal, etc. When the instructions in the storage medium are executed by the processor of the terminal, the terminal is able to perform the methods shown in the embodiments or combinations thereof.

[0244] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0245] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0246] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0247] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0248] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information retrieval method, characterized in that, The information query method includes: In response to a user's input of a motion-related query command, the query command is input into a first preset model to obtain vectorized first information; Determine the vectorized second information that matches the first information; The first information and the second information are input into the second preset model to obtain the first query result, which is used to guide the user's exercise. Send the first query result to the user.

2. The information query method according to claim 1, characterized in that, The first preset model includes a large sensor model and an alignment model; the step of inputting the query command into the first preset model to obtain vectorized first information includes: The query command is input into the sensor large model to obtain vectorized third information; The third information is input into the alignment model to obtain the first information in natural language vectorization.

3. The information query method according to claim 2, characterized in that, The first preset model further includes multiple linear detection layers, each corresponding to a sensor; the step of inputting the query command into the sensor large model to obtain vectorized third information includes: When the type of the query instruction is the first type, the query instruction is input into the sensor large model to obtain the third information, where the first type is a type related to motion guidance; When the type of the query instruction is the second type, the first motion information collected by the corresponding sensor is determined according to the query instruction, where the second type is a type related to information collection; The first motion information is input into the sensor large model and passed through the corresponding linear detection layer to obtain the third information.

4. The information query method according to claim 3, characterized in that, The step of inputting the first information and the second information into the second preset model to obtain the first query result includes: When the type of the query instruction is the first type, the first information and the second information are input into the second preset model to obtain the first query result; The information query method also includes: When the type of the query instruction is the second type, the first information is input into the second preset model to obtain the second query result; The second query result is sent to the user.

5. The information query method according to claim 2, characterized in that, The information query method also includes: Acquire fourth information from multiple sensors in different electronic devices; The discrete fourth pieces of information are converted into a series of consecutive fifth pieces of information; The sensor large model is obtained by using the fifth information described above for model training.

6. The information query method according to claim 5, characterized in that, The step of converting the discrete fourth pieces of information into a plurality of consecutive fifth pieces of information includes: The fourth information collected by each sensor and the corresponding label information of the sensor are mapped as a set of information to a high-dimensional vector space to obtain multiple sets of sixth information; Each set of the sixth information is serialized to obtain multiple sets of seventh information; After integrating the seventh information from each group, dimensionality reduction processing is performed to obtain the fifth information from each group.

7. The information query method according to claim 6, characterized in that, The process of serializing each group of the sixth information yields multiple groups of seventh information, including: Different objective functions are used to positionally encode the sixth information of the corresponding group to obtain the seventh information of each group; Wherein, the frequency parameter of the objective function is the acquisition frequency of the sensor corresponding to the sixth information.

8. The information query method according to claim 5, characterized in that, The process of training the model using the fifth information to obtain the large sensor model includes: The sensor large model is obtained by using a portion of the fifth information as the training set and another portion of the fifth information as the validation set for model training.

9. The information query method according to claim 5, characterized in that, After training the model using the fifth information to obtain the large sensor model, the information query method further includes: Multiple linear detection layers are constructed, each linear detection layer corresponding to a sensor; The corresponding linear detection layer is trained using the fourth information and the sensor large model.

10. The information query method according to claim 5, characterized in that, After training the model using the fifth information to obtain the large sensor model, the information query method further includes: Each of the fourth pieces of information is input into the sensor large model to obtain the vectorized eighth pieces of information; The alignment model is obtained by training the model using the eighth piece of information and the natural language information corresponding to the eighth piece of information.

11. The information query method according to claim 1, characterized in that, The second preset model is a large language model; the step of inputting the first information and the second information into the second preset model to obtain the first query result includes: Based on the first information and the second information, construct a prompt template; The prompt template is input into the large language model to obtain the first query result described in natural language text.

12. The information query method according to any one of claims 1 to 11, characterized in that, The information query method also includes: Collect the user's second motion information; The second motion information is input into the first preset model to obtain at least a portion of the information in the vector database; and / or, To acquire knowledge and information related to sports; Inputting the knowledge information into the first preset model yields at least a portion of the information in the vector database; and / or, Input the historical query results into the first preset model to obtain at least some information from the vector database; The second information is information from the vector database.

13. An information query device, characterized in that, The information query device includes: The first processing module is configured to respond to a motion-related query command input by a user, input the query command into a first preset model, and obtain vectorized first information. A determining module, configured to determine vectorized second information that matches the first information; The second processing module is configured to input the first information and the second information into a second preset model to obtain a first query result. The issuing module is configured to issue the first query result to the user.

14. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the information query method as described in any one of claims 1 to 12.

15. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the terminal, the terminal is able to perform the information query method as described in any one of claims 1 to 12.

Images