A personalized reminder management method and system for smart alarm clocks based on voice commands

Abstract

This application provides a personalized reminder management method and system for smart alarm clocks based on voice commands, belonging to the field of intelligent management technology. The method involves extracting multiple command entities from voice commands; constructing a dynamic knowledge graph based on all command entities; adding all command entities to a reminder scheme library; when the user inputs a new voice command, the entities in the new voice command are matched and filtered using a multi-level pipeline, resulting in multiple matching entities and newly added entities; change detection is performed on all newly added entities and incremental records in the new voice commands, and the knowledge graph and reminder scheme library are updated; based on the matching degree of each scheme branch in the updated reminder scheme library, the target scheme branch is selected, thereby generating personalized reminders for the user's smart alarm clock. Using the solution of this application, multiple voice commands containing aliases and ambiguous expressions can be accurately identified, and the reminder scheme of the smart alarm clock can be adjusted according to conditional relationships.

Description

Technical Field

[0001] This application relates to the field of intelligent management technology, and more specifically, to a personalized reminder management method and system for intelligent alarm clocks based on voice commands. Background Technology

[0002] With the popularization of smart home devices, smart alarm clocks have evolved from traditional timed ringing tools into personal assistant devices that integrate voice interaction and schedule management; users' personalized and dynamically changing reminder needs place higher demands on the semantic understanding, memory and adaptive capabilities of alarm clock systems.

[0003] Current smart alarm clock systems primarily use static rule matching or single semantic models to parse voice commands, which cannot effectively recognize different ways users describe the same event or handle ambiguous expressions. Secondly, they lack dynamic memory and online update mechanisms for users' historical reminder rules. When users add conditions or modify existing reminders, the new conditions cannot be integrated with existing rules; they are treated as independent commands, leading to fragmented reminder logic and invalid condition triggers. Therefore, accurately identifying multiple voice commands containing aliases and ambiguous expressions, and adjusting the smart alarm clock's reminder scheme based on conditional relationships, has become a major challenge for the industry. Summary of the Invention

[0004] This application provides a personalized reminder management method and system for smart alarm clocks based on voice commands, which can accurately identify multiple voice commands containing aliases and ambiguous expressions, and adjust the reminder scheme of the smart alarm clock according to the conditional relationship.

[0005] Firstly, this application provides a personalized reminder management method for a smart alarm clock based on voice commands, comprising the following steps: Collect user voice commands to the smart alarm clock and extract multiple command entities from the voice commands; A dynamic knowledge graph is built based on all instruction entities, and all instruction entities are added to a user-personalized tree-structured reminder solution library; When the user inputs a new voice command again, the entities in the new voice command are matched and filtered based on a multi-level pipeline to obtain multiple matching entities and new entities. Change detection is performed on all new entities and incremental records in the new voice command, and the knowledge graph and the reminder scheme library are updated. The target solution branch is selected based on the matching degree of each solution branch in the updated reminder solution library. All matching entities are added to the target solution branch, thereby generating personalized reminders for the user's smart alarm clock.

[0006] In some embodiments, constructing a dynamic knowledge graph based on all instruction entities specifically includes: Treat each instruction entity as a node and set the corresponding node attributes; Establish relation edges based on the semantic associations between various instruction entities; A dynamic knowledge graph is constructed based on all nodes and relation edges, and an initial embedding vector is generated for each node.

[0007] In some embodiments, adding all instruction entities to a user-personalized, tree-structured reminder scheme library specifically includes: Create an empty tree structure with pseudo-nodes as the root node, as the initial reminder scheme library for user personalization; Each instruction entity is sequentially attached to the root node as a child node according to its event dependency relationship, forming multiple initial branches; Set an initial priority weight for each branch and establish a bidirectional association index between each branch and all nodes in the knowledge graph.

[0008] In some embodiments, performing multi-level pipeline-based matching filtering on entities in the new voice command to obtain multiple matching entities and newly added entities specifically includes: Obtain all candidate entities extracted from the new voice command, and pre-set multiple priority matching pipelines; Multi-level matching is performed on all candidate entities based on the priority order of all matching pipelines and the knowledge graph. In each level of matching, all candidate entities are compared with all instruction entities in the knowledge graph; If a match is successful, the corresponding candidate entity is marked as a matched entity, the matching level is recorded, and matching for subsequent levels is stopped. If a match fails, the corresponding candidate entity will proceed to the next level of matching. Candidate entities that fail to match after multi-level matching are marked as new entities, resulting in multiple matched entities and new entities.

[0009] In some embodiments, performing change detection on incremental records in all newly added entities and new voice commands, and updating the knowledge graph and the reminder scheme library specifically includes: By using a change detector, all newly added entities are semantically similar to all nodes in the knowledge graph to obtain multiple changed related entities; Determine the relationships between entities in the new voice command, detect and mark all entity relationships, and obtain multiple incremental records; The incremental updater adds all change-related entities and all incremental records to the knowledge graph and updates the embedding vectors of all affected nodes. Add all changed related entities as new nodes to the branch of the aforementioned alert scheme library according to event dependencies; The conditional triggering logic for creating new branches or adjusting existing branches in the alert scheme library based on all incremental records.

[0010] In some embodiments, filtering target solution branches based on the matching degree of each solution branch in the updated reminder solution library specifically includes: Obtain the environmental parameters of the smart alarm clock at the current moment to get the context environment vector; Extract all nodes from the updated reminder scheme library to obtain the semantic sequence of each branch; Determine the attention weights of the context vector and each semantic sequence to obtain the matching degree of each branch; The branch with the highest matching degree is identified as the target solution branch, and the priority of all branches is readjusted based on the matching degree of the target solution branch.

[0011] In some embodiments, adding all matching entities to the target scheme branch to generate personalized reminders for the user's smart alarm clock specifically includes: Each matching entity is sequentially attached to the corresponding position in the target solution branch according to its corresponding semantic role, forming the updated reminder solution; The updated reminder scheme is serialized and encoded to obtain the reminder decision vector; The reminder decision vector is fused with the associated information in the updated knowledge graph, and then input into the text generation model to output personalized reminders for the user's smart alarm clock.

[0012] Secondly, this application provides a personalized reminder management system for smart alarm clocks based on voice commands, including: The acquisition module is used to collect the user's voice commands to the smart alarm clock and extract multiple command entities from the voice commands. The processing module is used to build a dynamic knowledge graph based on all instruction entities and add all instruction entities to the user's personalized tree-structured reminder solution library; The processing module is also used to perform multi-level pipeline-based matching and filtering on the entities in the new voice command when the user inputs a new voice command again, to obtain multiple matching entities and new entities, to perform change detection on all new entities and incremental records in the new voice command, and to update the knowledge graph and the reminder scheme library. The execution module is used to filter target solution branches based on the matching degree of each solution branch in the updated reminder solution library, add all matching entities to the target solution branch, and then generate personalized reminders for the user's smart alarm clock.

[0013] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the computer device executes the above-described intelligent alarm clock personalized reminder management method based on voice commands.

[0014] Fourthly, this application provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the aforementioned personalized reminder management method for a smart alarm clock based on voice commands.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application provides a personalized reminder management method and system for smart alarm clocks based on voice commands. First, it collects the user's voice commands to the smart alarm clock and extracts multiple command entities from these commands. A dynamic knowledge graph is constructed based on all command entities, and all command entities are added to a user-personalized tree-structured reminder scheme library. When the user inputs a new voice command, the entities in the new voice command are matched and filtered using a multi-level pipeline to obtain multiple matching entities and newly added entities. Change detection is performed on all newly added entities and incremental records in the new voice commands, and the knowledge graph and the reminder scheme library are updated. Based on the matching degree of each scheme branch in the updated reminder scheme library, a target scheme branch is selected, and all matching entities are added to the target scheme branch, thereby generating a personalized reminder for the user's smart alarm clock.

[0016] Therefore, in the process of a personalized reminder management method for a smart alarm clock based on voice commands, this application first collects the user's voice commands to the smart alarm clock and extracts multiple command entities from the voice commands; a dynamic knowledge graph is constructed based on all command entities, and all command entities are added to the user's personalized tree-like reminder scheme library; wherein, the reminder scheme library is a tree-like data structure that stores the user's personalized reminder rules, used to organize various reminder entities set by the user into a multi-level branch structure according to event dependencies, so that when the user adds conditional commands later, the target branch can be quickly located and the branch expanded through the tree structure; secondly, when the user inputs a new voice command again, the entities in the new voice command are matched and filtered based on a multi-level pipeline to obtain multiple matching entities and new entities, change detection is performed on all new entities and incremental records in the new voice commands, and the knowledge graph and the reminder scheme library are updated; wherein, the new entity is a marker that identifies new entities that have not yet been recorded in the knowledge graph, which helps to solve the problem of knowledge redundancy or omission caused by the inability to distinguish between new events in user commands and variant expressions of existing events. The above solution can accurately identify multiple voice commands containing aliases and vague expressions, and adjust the reminder scheme of the smart alarm clock according to the conditional relationship. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a voice-command-based smart alarm clock personalized reminder management method according to some embodiments of this application; Figure 2 This is a schematic diagram of the structure of the reminder scheme library according to some embodiments of this application; Figure 3 This is a schematic diagram of the structure of a dynamic knowledge graph according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a voice command-based smart alarm clock personalized reminder management system according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements a voice command-based smart alarm clock personalized reminder management method according to some embodiments of this application. Detailed Implementation

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a voice-command-based smart alarm clock personalized reminder management method according to some embodiments of this application. The voice-command-based smart alarm clock personalized reminder management method mainly includes the following steps: In step 101, the user's voice commands to the smart alarm clock are collected, and multiple command entities are extracted from the voice commands.

[0020] In specific implementation, the acquisition of user voice commands to the smart alarm clock can be achieved in the following way: An array of multiple microphones is built into the smart alarm clock device. This array can capture sound signals emitted by the user from different directions. When the user issues a voice command, the microphone array continuously monitors the surrounding environment, converts sound waves into electrical signals, and uses these electrical signals as the original voice signal. Next, the acquired original voice signal is input to an endpoint detection module. This module calculates the short-time energy and zero-crossing rate of the signal to distinguish between voice segments and silence segments. When the detected signal energy exceeds a preset threshold, it is determined that the user has started speaking, and recording begins. When the signal energy remains below the threshold for a preset duration, it is determined that speaking has ended, and recording stops. The recorded continuous signal segments are extracted as valid voice segments, and these voice segments are used as voice commands. The preset threshold can be set to 0.3 times the maximum amplitude of the voice signal. Other embodiments may also employ other methods, which are not limited here.

[0021] In some embodiments, extracting multiple instruction entities from the voice command can be achieved using the following steps: The voice command is input into a pre-trained language model to obtain the semantic embedding of each word. A relational graph convolutional network is used to encode all semantic embeddings, generating multiple entity-level representations; Decode time entities and event entities from all entity-level representations, and label the category and boundaries of each entity to obtain multiple instruction entities.

[0022] In specific implementation, inputting the voice command into a pre-trained language model to obtain the semantic embedding of each word can be achieved as follows: the voice command is first converted into a text sequence by a speech recognition module, and then the text sequence is input into the pre-trained language model. This language model uses a Transformer architecture and encodes the text sequence through a multi-layer self-attention mechanism, generating a fixed-dimensional vector for each word in the sequence; this vector is the semantic embedding of that word. Encoding all semantic embeddings using a relational graph convolutional network to generate multiple entity-level representations can be achieved as follows: all word semantic embeddings are input into the relational graph convolutional network. This network constructs a graph structure based on the positional relationships and syntactic dependencies between words, aggregating information from adjacent nodes. The representation of each node is updated, and the final output is a vector sequence of the same length as the input sequence. Each vector in this sequence is the entity-level representation of the corresponding word. Time entities and event entities are decoded from all entity-level representations, and the category and boundary of each entity are labeled to obtain multiple instruction entities. This can be achieved as follows: all entity-level representations are input into a conditional random field decoder, which predicts entity category labels and boundary labels for each position in the sequence. Entity categories include time categories and event categories, and boundary labels identify the start and end positions of the entity. All text fragments predicted as entities and their categories are extracted, with each fragment as an instruction entity. All instruction entities are used as the final extraction result to obtain multiple instruction entities. Other embodiments may also use other methods to achieve this, which are not limited here.

[0023] It should be noted that the semantic embedding in this application is used to represent the vectorized features of each word in the semantic space; the entity-level representation is used to reflect the semantic features of each entity; and the instruction entity is a structured information unit used to construct the knowledge graph and the reminder scheme library.

[0024] In step 102, a dynamic knowledge graph is constructed based on all instruction entities, and all instruction entities are added to the user's personalized tree-structured reminder scheme library.

[0025] In some embodiments, constructing a dynamic knowledge graph based on all instruction entities can be achieved using the following steps: Treat each instruction entity as a node and set the corresponding node attributes; Establish relation edges based on the semantic associations between various instruction entities; A dynamic knowledge graph is constructed based on all nodes and relation edges, and an initial embedding vector is generated for each node.

[0026] In specific implementation, treating each instruction entity as a node and setting corresponding node attributes can be achieved in the following way: Traverse all instruction entities, create a graph node for each entity, and set attribute information for that node. Attribute information should at least include entity type, entity text content, and the timestamp when the entity was collected. Use all created nodes as the initial node set of the knowledge graph. Establishing relation edges based on the semantic associations between instruction entities can be achieved in the following way: First, obtain a preset set of semantic relation types, including temporal relations, conditional relations, causal relations, and adjoint relations. For each instruction entity, extract its dependency syntax structure in the speech command, and identify the core predicates and modification relations between entities using a dependency parser. When two entities are governed by the same predicate, ... The relation type is determined based on the semantics of the predicate. For example, for the entities "get up" and "7 o'clock," if dependency analysis shows that "7 o'clock" modifies "get up" as a time adverb, then a temporal relation edge is established from "get up" to "7 o'clock," and the relation type is labeled "triggered by." For the entities "rain" and "postpone," if the instruction contains a conditional structure "postpone if it rains," then a conditional relation edge is established from "rain" to "postpone," and the relation type is labeled "conditionally triggered." All directed edges corresponding to the identified semantic associations are used as the initial edge set of the knowledge graph. A dynamic knowledge graph is constructed based on all nodes and relation edges, and an initial embedding vector is generated for each node. This can be achieved by combining all nodes and all relation edges to form an initial knowledge graph structure, as shown in the reference. Figure 3 As shown, this diagram is a schematic representation of the structure of a dynamic knowledge graph in some embodiments of this application. The diagram includes nodes corresponding to each entity and edges constructed from the relationship types between entities. A graph neural network is used to encode this knowledge graph. By aggregating the neighbor node information of each node, an initial embedding vector with a fixed dimension is generated for each node. This fixed dimension consists of the following four aspects: node type dimension, used to indicate whether the entity belongs to the time category, event category, or condition category; node content dimension, used to indicate the semantic features of the entity text; node relationship dimension, used to indicate the edge type distribution between the node and its neighboring nodes; and node attribute dimension, used to indicate additional attributes such as the entity's timestamp and frequency of occurrence. This vector is then used as the initial semantic feature representation of the node. Other embodiments may also use other methods to implement this, which are not limited here.

[0027] It should be noted that the relation edges in this application are structures used to represent semantic connections between nodes in a knowledge graph; the initial embedding vector is a numerical representation of the initial semantic features of a node in the knowledge graph. After the knowledge graph is constructed, each node is assigned an initial vector that integrates node type, node content, node relation distribution, and node attribute dimensions. This facilitates the rapid location of candidate entities through vector similarity calculation during subsequent entity matching. At the same time, it provides basic features for the semantic encoding of branch nodes in the tree-like reminder scheme library, solving the problem of insufficient understanding of entity semantics in user commands by existing alarm clock systems.

[0028] In some embodiments, adding all instruction entities to a user-personalized, tree-structured reminder scheme library can be achieved through the following steps: Create an empty tree structure with pseudo-nodes as the root node, as the initial reminder scheme library for user personalization; Each instruction entity is sequentially attached to the root node as a child node according to its event dependency relationship, forming multiple initial branches; Set an initial priority weight for each branch and establish a bidirectional association index between each branch and all nodes in the knowledge graph.

[0029] In practical implementation, creating an empty tree structure with a pseudo-node as the root node, serving as the initial personalized reminder scheme library for the user, can be implemented as follows: Create a tree data structure in memory, setting a virtual node as the root node of the entire tree. This root node does not carry any entity information; it only serves as the starting anchor point of the tree structure. This tree structure serves as the initial empty reminder scheme for subsequent attachment of user reminder entities. Attaching each instruction entity as a child node under the root node according to event dependencies, forming multiple initial branches, can be implemented as follows: Traverse all extracted instruction entities, identify their event dependencies, organize entities with dependencies according to their dependency order, making the first event entity a child node of the root node, and subsequent dependent events children nodes of that child node, and so on, forming a multi-layered path starting from the root node; (See reference...) Figure 2As shown, this figure is a schematic diagram of the structure of the reminder solution library in some embodiments of this application. The figure contains multiple entities, represented by circles, and the dependencies derived from voice commands, represented by arrows. All paths are used as initial branches of the reminder solution library. Setting an initial priority weight for each branch and establishing a bidirectional association index between each branch and all nodes in the knowledge graph can be achieved in the following way: assigning an initial weight value to each initial branch, which represents the initial probability of the branch being selected; simultaneously, establishing an index relationship between each node in the branch and the corresponding entity node in the knowledge graph, through which the entity information corresponding to the branch node can be quickly located in the knowledge graph; setting the weights... The index relationship is stored as an additional attribute of the branch; the weight value is set in the following way: First, the total frequency of all entities in the branch appearing in the user's historical voice commands is counted. The higher the frequency, the higher the weight value. Second, the number of dependencies contained in the branch is calculated. The more dependencies, the more complex the reminder logic of the branch is and the more likely it is to be needed by the user, so the weight is increased accordingly. Third, the specificity of the time entities in the branch is evaluated. Time accurate to the minute receives a higher weight bonus than vague time periods. Finally, the above factors are weighted and summed, and the sum is normalized to the range of 0 to 1 as the initial weight value of the branch. Other implementation methods can also be used in other embodiments, which are not limited here.

[0030] It should be noted that the reminder scheme in this application is a tree-structured data structure used to store personalized reminder rules. This structure defines the organization of reminder logic from the perspective of personalized user memory, organizing various reminder entities set by the user into a multi-level branch structure according to event dependencies. This facilitates quick location of the target branch and branch expansion when the user adds subsequent conditional instructions. It also supports dynamically calculating the matching degree of each branch based on environmental parameters, solving the problem that existing alarm clock systems cannot remember and manage personalized reminder rules. The initial branch is a tree-structured path representing the execution path of the user reminder event, used to link instructions with temporal or conditional dependencies. The entity organization represents the complete path from the root node to the leaf node, facilitating the selection of the most suitable reminder scheme based on branch matching degree when the user inputs conditional commands. The association index is a connection identifier used to establish the mapping relationship between the reminder scheme library branches and knowledge graph nodes. When a branch node is updated, the index can be used to quickly locate the corresponding entity node in the knowledge graph and obtain its embedding vector. At the same time, after the knowledge graph is updated, the index can be used to reverse locate the affected reminder scheme branches and update them synchronously. This facilitates bidirectional linkage and consistency maintenance between the knowledge graph and the reminder scheme library, solving the problem of asynchronous knowledge updates and reminder rules when user habits change in existing alarm clock systems.

[0031] In step 103, when the user inputs a new voice command again, the entities in the new voice command are matched and filtered based on a multi-level pipeline to obtain multiple matching entities and newly added entities. Change detection is performed on all newly added entities and incremental records in the new voice command, and the knowledge graph and the reminder scheme library are updated.

[0032] In some embodiments, performing multi-level pipeline-based matching filtering on entities in the new voice command to obtain multiple matching entities and newly added entities can be achieved through the following steps: Obtain all candidate entities extracted from the new voice command, and pre-set multiple priority matching pipelines; Multi-level matching is performed on all candidate entities based on the priority order of all matching pipelines and the knowledge graph. In each level of matching, all candidate entities are compared with all instruction entities in the knowledge graph; If a match is successful, the corresponding candidate entity is marked as a matched entity, the matching level is recorded, and matching for subsequent levels is stopped. If a match fails, the corresponding candidate entity will proceed to the next level of matching. Candidate entities that fail to match after multi-level matching are marked as new entities, resulting in multiple matched entities and new entities.

[0033] In specific implementation, obtaining all candidate entities extracted from the new voice command and pre-setting multiple priority matching pipelines can be achieved in the following way: perform the same entity extraction operation as the previous voice command on the newly input voice command to obtain a set of candidate entities; pre-define five matching pipelines, arranged in order of matching accuracy from high to low, namely, identifier matching pipeline, exact name matching pipeline, type and time attribute matching pipeline, alias and synonym matching pipeline, and fuzzy string matching pipeline. These five pipelines serve as the execution units for matching and filtering. Multi-level matching of all candidate entities based on the priority order of all matching pipelines and the knowledge graph can be implemented as follows: For each candidate entity, matching is performed sequentially in the order of identifier pipeline, exact name pipeline, type and time attribute pipeline, alias and synonym pipeline, and fuzzy string pipeline. During matching, each pipeline compares the candidate entity with the corresponding attributes of all entities in the knowledge graph to determine if the matching conditions of that pipeline are met. In each level of matching, comparing all candidate entities with all instruction entities in the knowledge graph can be implemented as follows: When executing a certain level of matching pipeline, the candidate entity is compared with each existing entity stored in the knowledge graph using the comparison operation corresponding to that pipeline. For example, when executing the identifier matching pipeline, the unique identifier of the candidate entity is precisely matched with the identifier field of the knowledge graph entity; when executing the exact name matching pipeline, the text of the candidate entity is compared with the name field of the knowledge graph entity for string equality; when executing the type and time attribute matching pipeline... The process involves comparing the type label and time attribute value of the candidate entity with the type and time fields of the knowledge graph entity one by one. When executing the alias and synonym matching pipeline, the candidate entity text is compared with each item in the alias list of the knowledge graph entity for string equality. When executing the fuzzy string matching pipeline, the edit distance algorithm is used to calculate the similarity between the candidate entity text and the name field of the knowledge graph entity. A successful match is determined when the similarity exceeds a preset threshold. If a match is successful, the corresponding candidate entity is marked as a matched entity, the matching level is recorded, and matching at subsequent levels is stopped. If a match fails, the corresponding candidate entity is moved to the next level of matching. This can be implemented as follows: when the candidate entity successfully matches an entity in the knowledge graph at a certain pipeline level, the candidate entity is marked as a matched entity, the current pipeline level (e.g., level 1 to level 5) is recorded, and subsequent pipeline matching for the candidate entity is immediately terminated. If the current pipeline matching fails, the candidate entity is passed to the next priority pipeline for further matching. Other implementation methods can also be used in other embodiments, which are not limited here.

[0034] In specific implementation, candidate entities that fail to match after multi-level matching are marked as new entities. The multiple matched entities and new entities can be obtained in the following way: when any candidate entity fails to match any entity in the knowledge graph after matching through all five matching pipelines, the candidate entity is marked as a new entity; all matched entities and all marked new entities are used together as the output of the matching filter; other methods can also be used in other embodiments, which are not limited here.

[0035] It should be noted that the matching pipeline in this application consists of multiple processing units used to perform entity matching in priority order. It is a hierarchical definition of the entity screening process from the perspective of multi-level matching. It is used to compare candidate entities step by step according to the priority order of identifier matching, exact name matching, type and time attribute matching, alias and synonym matching, and fuzzy string matching. This makes it easier to accurately identify entities corresponding to existing entities in the knowledge graph when the user inputs voice commands containing aliases, synonyms, or fuzzy expressions, through a multi-level progressive matching strategy. This improves the accuracy of parsing different statements about the same event and fuzzy expressions. Multi-level matching is a hierarchical matching process used to screen candidate entities step by step. The matching level is a record value used to identify which level of the pipeline the candidate entity was successfully matched at.

[0036] In some embodiments, the following steps can be used to perform change detection on incremental records in all newly added entities and new voice commands, and to update the knowledge graph and the reminder scheme library: By using a change detector, all newly added entities are semantically similar to all nodes in the knowledge graph to obtain multiple changed related entities; Determine the relationships between entities in the new voice command, detect and mark all entity relationships, and obtain multiple incremental records; The incremental updater adds all change-related entities and all incremental records to the knowledge graph and updates the embedding vectors of all affected nodes. Add all changed related entities as new nodes to the branch of the aforementioned alert scheme library according to event dependencies; The conditional triggering logic for creating new branches or adjusting existing branches in the alert scheme library based on all incremental records.

[0037] In specific implementation, the change detector compares the semantic similarity of all newly added entities with all nodes in the knowledge graph to obtain multiple change-related entities. This can be achieved in the following way: all marked newly added entities are input into the change detector, which calculates the semantic similarity between the text of each new entity and the name, alias, and attribute value of each node in the knowledge graph, using cosine similarity as the metric. When the similarity is lower than a preset threshold, it is determined that the entity does not have a true corresponding node in the knowledge graph and is marked as a change-related entity; when the similarity is higher than the threshold, it is determined that the entity is a variant expression of an existing entity, and an association is established between the entity and the corresponding node, but it is not added as a new node. The preset threshold is set to 0.7, which means that when the cosine value of the angle between the semantic vectors of two texts exceeds 0.7, they are considered to be highly semantically similar. This value is determined based on the statistical distribution of semantic similarity of texts in a general domain and can be adjusted according to actual needs. The data distribution of the application scenario is dynamically adjusted; the relationships between entities in the new voice command are determined, and all relationships between entities are detected and marked to obtain multiple incremental records. This can be achieved in the following way: parsing the text structure of the new voice command and identifying the logical relationships expressed therein. For example, for the command "If it rains, postpone for half an hour", dependency parsing identifies "if" as a conditional conjunction leading a conditional clause, "rain" as the core predicate of the clause, "postpone" as the core predicate of the main clause, and "then" as the connection relationship between the condition and the result. Based on this, it is parsed that there is a conditional triggering relationship between "rain" and "postpone". For each identified relationship, it is compared with the relationship types that already exist in the knowledge graph. If the relationship type has not yet been recorded in the knowledge graph, it is marked as an incremental record, and the head entity and tail entity associated with the relationship are recorded. Other embodiments can also be implemented in other ways, which are not limited here.

[0038] In specific implementation, adding all changed entities and all incremental records to the knowledge graph through an incremental updater and updating the embedding vectors of all affected nodes can be achieved in the following way: adding all marked changed entities as new nodes to the knowledge graph and adding all marked incremental records as new edges to the knowledge graph; generating initial embedding vectors for newly added nodes; recalculating the embedding vectors of existing nodes affected by new nodes and edges using a graph neural network, updating them by aggregating information from their neighboring nodes to obtain a new representation and replacing the original embedding vectors with this new representation; adding all changed entities as new nodes to the branches of the reminder scheme library according to sub-event dependencies can be achieved in the following way: traversing all marked changed entities and determining the event type of the entity; identifying its dependencies on existing entities in the reminder scheme library based on the event type. For example, the conditional event "rain" should be mounted as a dependent condition node for the "wake up" event; the new node is added as a child node to the corresponding position of the corresponding branch to form an expanded branch path; the condition triggering logic of creating new branches or adjusting existing branches in the reminder scheme library based on all incremental records can be implemented in the following way: for each marked incremental record, find the branch where the head entity and tail entity associated with the relationship are located in the reminder scheme library; if the relationship represents a new triggering condition, create a new branch under the head entity node and make the tail entity a child node of the branch; if the relationship is a modification of the existing condition, adjust the triggering condition of the corresponding node in the existing branch, for example, update the triggering condition of the "7 o'clock" node to "7 o'clock and no rain"; save the created or adjusted branch structure to the reminder scheme library; other embodiments can also be implemented in other ways, which are not limited here.

[0039] It should be noted that the "new entity" in this application is a marker used to identify new entities that have not yet been recorded in the knowledge graph; the "change association" is a filtering result used to determine the truly new entities that need to be added to the knowledge graph; the filtering result used to determine the truly new entities that need to be added to the knowledge graph helps to avoid duplicate variant nodes in the knowledge graph, while ensuring that truly new information can be included in a timely manner, solving the problem of knowledge redundancy or omission caused by the inability to distinguish between completely new events in user commands and variant expressions of existing events; the "incremental record" is a marker used to identify new relation types that have not yet been recorded in the knowledge graph, which facilitates the dynamic expansion of the relation type set of the knowledge graph, enabling the alarm clock system to understand the conditional commands subsequently added by the user and establish corresponding triggering logic, solving the problem that existing alarm clock systems cannot handle condition-dependent reminders; the "event dependency relationship" is the logical basis used to determine the mounting position of new nodes in the reminder scheme library; the "conditional triggering logic" is the judgment rule used to determine when the reminder scheme is activated.

[0040] It should be noted that subsequent user input often supplements or modifies previous reminders. For example, in the phrase "If it rains, change it to 7:30," the conditional relationship "rains" and the modification action "change to" do not exist in the initial knowledge graph. This solution first uses a multi-level matching mechanism to perform hierarchical filtering of candidate entities in new instructions, ensuring accurate recognition even if the voice input contains aliases or ambiguous expressions. Then, a change detector compares the semantic similarity of unmatched new entities with existing nodes in the graph, distinguishing between changed related entities and variant expressions. Finally, an incremental updater synchronously updates the changed related entities and incremental records to the knowledge graph and reminder scheme library. By adopting the above steps, online adaptive memorization of personalized reminder rules can be achieved, avoiding redundancy in the knowledge graph. This enables the alarm clock system to understand multiple condition-dependent instructions and dynamically expand the branch structure of the reminder scheme library, effectively solving the problem that existing alarm clocks cannot handle user-added conditions or modifications to existing reminders.

[0041] In step 104, the target solution branch is selected based on the matching degree of each solution branch in the updated reminder solution library, and all matching entities are added to the target solution branch to generate personalized reminders for the user's smart alarm clock.

[0042] In some embodiments, filtering target solution branches based on the matching degree of each solution branch in the updated reminder solution library can be achieved by the following steps: Obtain the environmental parameters of the smart alarm clock at the current moment to get the context environment vector; Extract all nodes from the updated reminder scheme library to obtain the semantic sequence of each branch; Determine the attention weights of the context vector and each semantic sequence to obtain the matching degree of each branch; The branch with the highest matching degree is identified as the target solution branch, and the priority of all branches is readjusted based on the matching degree of the target solution branch.

[0043] In specific implementation, obtaining the environmental parameters of the smart alarm clock at the current moment and obtaining the context environment vector can be achieved in the following way: at the alarm trigger time, the current environmental parameters are obtained through sensors, including the current time, weather conditions, user location, and user response behavior to historical reminders; these parameters are converted into numerical form, and all values ​​are concatenated into a fixed-dimensional vector as the context environment vector; extracting all nodes in the updated reminder scheme library and obtaining the semantic sequence of each branch can be achieved in the following way: traverse each branch in the reminder scheme library, starting from the root node, extract the entity embedding vector corresponding to each node in the order from root to leaf node, and arrange these vectors according to the order of the branch path to form a vector sequence. Each branch corresponds to a vector sequence, which serves as the semantic sequence of that branch. The matching degree of each branch can be determined by assigning attention weights to the context vector and each semantic sequence, as follows: For each branch's semantic sequence, calculate the dot product of each vector in the sequence with the context vector to obtain the attention score for each position; convert the attention scores of all positions into a weight distribution using exponential normalization; sum each vector in the semantic sequence according to its corresponding weight to obtain a comprehensive representation vector; calculate the cosine similarity between this comprehensive representation vector and the context vector, and this similarity value is the matching degree of that branch; the branch with the highest matching degree is determined as the target solution branch, and the priority order of all branches is readjusted based on the matching degree of the target solution branch, as follows: Compare the matching degree values ​​of all branches and select the branch with the highest matching degree as the target solution branch; for all branches, sort their matching degree values ​​in descending order, with the branch with the highest matching degree at the front, and so on, using the reordered order as the new priority order for each branch, used for the priority matching order of branches when a new instruction arrives; other methods can also be used in other embodiments, which are not limited here.

[0044] It should be noted that the context vector in this application is a numerical feature used to represent the current state of the external environment; the semantic sequence is a sequential representation of the semantic content of each node in the branch; the matching degree is a numerical indicator used to measure the degree of adaptation between the branch and the current environment; and the priority ranking is a ranking list used to determine the order in which branches are examined in the next matching.

[0045] In some embodiments, adding all matching entities to the target scheme branch to generate personalized reminders for the user's smart alarm clock can be achieved through the following steps: Each matching entity is sequentially attached to the corresponding position in the target solution branch according to its corresponding semantic role, forming the updated reminder solution; The updated reminder scheme is serialized and encoded to obtain the reminder decision vector; The reminder decision vector is fused with the associated information in the updated knowledge graph, and then input into the text generation model to output personalized reminders for the user's smart alarm clock.

[0046] In specific implementation, each matching entity is sequentially attached to the corresponding position in the target solution branch according to its corresponding semantic role to form the updated reminder scheme. This can be achieved in the following way: For each matching entity, determine its semantic role, including time role, event role, and condition role; determine the attachment position of the entity in the target solution branch according to the semantic role, for example, a time entity is attached to the time node position, and an event entity is attached to the action node position; add the matching entity as a new node to the corresponding position to form a complete reminder rule structure; serialize and encode the updated reminder scheme to obtain the reminder decision vector. This can be achieved in the following way: traverse the updated reminder scheme in order from the root node to the leaf node, extract the entity embedding vector of each node in turn, and concatenate these vectors into a long vector according to the path order; input this long vector into a multilayer perceptron network. The network outputs a fixed-dimensional vector through multi-layer linear transformations and non-linear activations; this vector is the reminder decision vector. The reminder decision vector is then fused with the associated information in the updated knowledge graph and input into a text generation model to output personalized reminders for the user's smart alarm clock. This can be achieved as follows: Supplementary information related to entities in the reminder decision vector is retrieved from the updated knowledge graph, such as user habit preferences corresponding to reminder events and coping strategies for conditional events. The reminder decision vector is then concatenated and fused with the embedding vectors of this supplementary information to obtain a comprehensive feature vector. This comprehensive feature vector is input into a pre-trained autoregressive text generation model, which generates natural language text word by word based on the feature vector. The generated text serves as the personalized reminder for the user's smart alarm clock. Other implementation methods can also be used in other embodiments, which are not limited here.

[0047] Furthermore, in another aspect of this application, in some embodiments, this application provides a personalized reminder management system for smart alarm clocks based on voice commands, see reference. Figure 4 The figure is a schematic diagram of the structure of a smart alarm clock personalized reminder management system according to some embodiments of this application. The smart alarm clock personalized reminder management system includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to collect the user's voice commands to the smart alarm clock and extract multiple command entities from the voice commands. Processing module 402, in this application, is used to construct a dynamic knowledge graph based on all instruction entities and add all instruction entities to the user's personalized tree-like reminder scheme library; It should be noted that the processing module 402 in this application is also used to perform multi-level pipeline-based matching filtering on the entities in the new voice command when the user inputs a new voice command again, to obtain multiple matching entities and new entities, to perform change detection on all new entities and incremental records in the new voice command, and to update the knowledge graph and the reminder scheme library. The execution module 403 in this application is mainly used to filter the target scheme branch based on the matching degree of each scheme branch in the updated reminder scheme library, add all matching entities to the target scheme branch, and then generate personalized reminders for the user's smart alarm clock.

[0048] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described intelligent alarm clock personalized reminder management method based on voice commands.

[0049] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a voice-command-based smart alarm clock personalized reminder management method according to some embodiments of this application. The voice-command-based smart alarm clock personalized reminder management method in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0050] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0051] The communication bus 502 can be used to transmit information between the aforementioned components.

[0052] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0053] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0054] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0055] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0056] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0057] In addition, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described personalized reminder management method for a smart alarm clock based on voice commands.

[0058] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0059] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A personalized reminder management method for a smart alarm clock based on voice commands, characterized in that, Includes the following steps: Collect user voice commands to the smart alarm clock and extract multiple command entities from the voice commands; A dynamic knowledge graph is built based on all instruction entities, and all instruction entities are added to a user-personalized tree-structured reminder solution library; When the user inputs a new voice command again, the entities in the new voice command are matched and filtered based on a multi-level pipeline to obtain multiple matching entities and new entities. Change detection is performed on all new entities and incremental records in the new voice command, and the knowledge graph and the reminder scheme library are updated. Based on the matching degree of each branch in the updated reminder solution library, the target solution branch is selected, all matching entities are added to the target solution branch, and then personalized reminders for the user's smart alarm clock are generated.

2. The method as described in claim 1, characterized in that, Building a dynamic knowledge graph based on all instruction entities specifically includes: Treat each instruction entity as a node and set the corresponding node attributes; Establish relation edges based on the semantic associations between various instruction entities; A dynamic knowledge graph is constructed based on all nodes and relation edges, and an initial embedding vector is generated for each node.

3. The method as described in claim 1, characterized in that, Adding all instruction entities to the user's personalized, tree-structured reminder scheme library specifically includes: Create an empty tree structure with pseudo-nodes as the root node, as the initial reminder scheme library for user personalization; Each instruction entity is sequentially attached to the root node as a child node according to its event dependency relationship, forming multiple initial branches; Set an initial priority weight for each branch and establish a bidirectional association index between each branch and all nodes in the knowledge graph.

4. The method as described in claim 1, characterized in that, The entities in the new voice command are subjected to multi-level pipeline-based matching and filtering to obtain multiple matching entities and newly added entities, specifically including: Obtain all candidate entities extracted from the new voice command, and pre-set multiple priority matching pipelines; Multi-level matching is performed on all candidate entities based on the priority order of all matching pipelines and the knowledge graph. In each level of matching, all candidate entities are compared with all instruction entities in the knowledge graph; If a match is successful, the corresponding candidate entity is marked as a matched entity, the matching level is recorded, and matching for subsequent levels is stopped. If a match fails, the corresponding candidate entity will proceed to the next level of matching. Candidate entities that fail to match after multi-level matching are marked as new entities, resulting in multiple matched entities and new entities.

5. The method as described in claim 1, characterized in that, The process of performing change detection on incremental records for all newly added entities and new voice commands, and updating the knowledge graph and the reminder scheme library, specifically includes: By using a change detector, all newly added entities are semantically similar to all nodes in the knowledge graph to obtain multiple changed related entities; Determine the relationships between entities in the new voice command, detect and mark all entity relationships, and obtain multiple incremental records; The incremental updater adds all change-related entities and all incremental records to the knowledge graph and updates the embedding vectors of all affected nodes. Add all changed related entities as new nodes to the branch of the aforementioned alert scheme library according to event dependencies; The conditional triggering logic for creating new branches or adjusting existing branches in the alert scheme library based on all incremental records.

6. The method as described in claim 1, characterized in that, Filtering target solution branches based on the matching degree of each solution branch in the updated reminder solution library specifically includes: Obtain the environmental parameters of the smart alarm clock at the current moment to get the context environment vector; Extract all nodes from the updated reminder scheme library to obtain the semantic sequence of each branch; Determine the attention weights of the context vector and each semantic sequence to obtain the matching degree of each branch; The branch with the highest matching degree is identified as the target solution branch, and the priority of all branches is readjusted based on the matching degree of the target solution branch.

7. The method as described in claim 1, characterized in that, Adding all matching entities to the target solution branch, and then generating personalized reminders for the user's smart alarm clock, specifically includes: Each matching entity is sequentially attached to the corresponding position in the target solution branch according to its corresponding semantic role, forming the updated reminder solution; The updated reminder scheme is serialized and encoded to obtain the reminder decision vector; The reminder decision vector is fused with the associated information in the updated knowledge graph, and then input into the text generation model to output personalized reminders for the user's smart alarm clock.

8. A personalized reminder management system for intelligent alarm clocks based on voice commands, characterized in that, include: The acquisition module is used to collect the user's voice commands to the smart alarm clock and extract multiple command entities from the voice commands. The processing module is used to build a dynamic knowledge graph based on all instruction entities and add all instruction entities to the user's personalized tree-structured reminder solution library; The processing module is also used to perform multi-level pipeline-based matching and filtering on the entities in the new voice command when the user inputs a new voice command again, to obtain multiple matching entities and new entities, to perform change detection on all new entities and incremental records in the new voice command, and to update the knowledge graph and the reminder scheme library. The execution module is used to filter target solution branches based on the matching degree of each solution branch in the updated reminder solution library, add all matching entities to the target solution branch, and then generate personalized reminders for the user's smart alarm clock.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to call and run the computer programs from the memory, so that the computer device executes the voice command-based smart alarm clock personalized reminder management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions or code that, when executed on a computer, cause the computer to implement the personalized reminder management method for smart alarm clocks based on voice commands as described in any one of claims 1 to 7.

Images