Intelligent assistant system and implementation method thereof

Abstract

The invention relates to the technical field of intelligent assistants, and discloses an intelligent assistant system and an implementation method thereof, the intelligent assistant system comprises a data acquisition unit, a preprocessing unit, a data analysis unit, a knowledge graph construction unit, an intelligent interaction unit and a personalized service unit; multi-source data collection is achieved, web pages, third-party APIs and user behavior data are covered, and richness, timeliness and pertinence of information are guaranteed. Through efficient data preprocessing, the data quality is improved; text content and emotional tendency are accurately mastered through multiple technologies in text analysis; user behavior analysis deeply excavates an association relationship, constructs a portrait and analyzes a trend; the knowledge graph construction effectively organizes knowledge and improves retrieval efficiency, personalized service provides personalized experience based on user portraits and behavior analysis, user satisfaction is improved on the whole, intelligent, convenient and personalized service experience is brought to users, and the method has wide application value and market potential.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent assistants, and in particular to an intelligent assistant system and an implementation method thereof. Background Art

[0002] In today's digital age, people's demand for information acquisition and processing is growing. As a tool that can provide users with convenient and efficient services, intelligent assistant systems have received widespread attention. With the popularization of the Internet and the explosive growth of data, users are facing the problem of information overload. How to accurately obtain the required information from massive data and provide personalized services has become a key challenge.

[0003] Traditional information retrieval methods can only provide simple keyword matching results, making it difficult to understand the user's true intentions and emotional tendencies, and unable to provide in-depth analysis and personalized recommendations. At the same time, there are also deficiencies in the integration and utilization of multi-source data, and it is impossible to fully explore the relationship and value between data.

[0004] In the field of natural language processing, although there are some technologies and methods, there are still many problems in practical applications, such as the accuracy of text analysis, the depth of user behavior analysis, and the perfection of knowledge graph construction. For intelligent interaction, how to achieve more natural and accurate semantic understanding and intention judgment, and how to make more accurate recommendations based on user portraits in personalized services are all urgent problems to be solved.

[0005] Therefore, in order to meet users' higher expectations and demands for intelligent assistant systems, improve the efficiency and quality of information processing, and achieve more intelligent and personalized services, the present invention proposes an intelligent assistant system and an implementation method thereof. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent assistant system and an implementation method thereof, which solves the technical problems raised in the background technology.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] An intelligent assistant system and an implementation method thereof, comprising:

[0009] A data collection unit, used for collecting multi-source data from multiple data sources;

[0010] The multi-source data includes various web page data obtained from web crawlers, and the web page data covers news websites, forums, blogs, etc. to obtain real-time news information and text data corresponding to user comments;

[0011] Access third-party APIs to obtain data in professional fields; and collect user behavior data when using smart assistants, including text data corresponding to search keywords, clicked content, browsing time, etc.

[0012] A preprocessing unit is used to clean the collected text data, wherein the text data includes but is not limited to HTML tags, advertising information, and special characters, and to filter and replace the text data by any one of regular expressions or string processing functions, so as to retain only meaningful text content;

[0013] The data analysis unit is used to perform multi-faceted analysis on the pre-processed multi-source data, including:

[0014] Text analysis: Use natural language processing technology for word segmentation, part-of-speech tagging and named entity recognition. Use the Jieba word segmentation tool to segment the text, use the NLTK library to perform part-of-speech tagging, and use a conditional random field-based method to perform named entity recognition; determine the sentiment tendency of the text, build a support vector machine classification model, collect text data with sentiment annotations as a training set and extract text features, train a support vector machine classification model to determine the sentiment category based on text features, extract text features of new texts and input them into the pre-trained support vector machine classification model for sentiment prediction to determine the sentiment tendency;

[0015] User behavior analysis: Use association rule mining algorithms to mine the associations between user behaviors, build user portraits based on users’ basic information and behavior characteristics, and use time series analysis methods to analyze the changing trends of user behavior data over time;

[0016] The knowledge graph construction unit is used to screen and determine the entities in the knowledge graph based on the results of named entity recognition in the data analysis unit, including creating node labels and relationship types, inserting entities and relationships into the graph database one by one according to the data format requirements to establish the graph structure of the knowledge graph and create indexes based on the name attributes of the entities, and importing the sorted entities and relationships into the graph database through the data import tool, using entities as nodes and relationships as edges, and storing the attributes of the entities and relationships in the form of attribute values ​​in the nodes and edges;

[0017] The intelligent interaction unit is used to perform text parsing on natural language text input by the user, including converting the user's questions into semantic representations that can be understood by computers through semantic parsing technology, classifying the user's questions and judging the user's intentions using a convolutional neural network model, generating answers based on the knowledge graph and data analysis results, and recording the previous conversation content and context information during the interaction with the user to answer based on the context; the personalized service unit is used to make personalized recommendations based on user portraits and behavior analysis results, including making recommendations based on the user's interest in specific types of content and combining the user's location information and time information, and using a collaborative filtering algorithm to make recommendations based on the similarity of user behavior.

[0018] As a further solution of the present invention: the data acquisition unit obtains web page data from the web crawler by: using Python's Scrapy framework to write a crawler program, defining a list of websites to be crawled and page rules, and regularly crawling web page content and storing it in a local database or data storage platform.

[0019] As a further solution of the present invention: the access to the third-party API is specifically:

[0020] By registering and calling a third-party API, you can send a request in the request format specified in the API document to obtain data in professional fields including temperature, weather conditions, and weather information corresponding to air quality.

[0021] As a further solution of the present invention: the collection of user behavior data is specifically: by embedding data collection code in the smart assistant application, when the user performs an operation, the relevant behavior data is automatically recorded and uploaded to the server for storage and analysis.

[0022] As a further solution of the present invention: in the cleaning process of the text data in the preprocessing unit, when regular expressions are used, the specific expression pattern and matching rules are set and adjusted according to actual needs to achieve effective filtering and replacement of HTML tags, advertising information, and special characters.

[0023] As a further solution of the present invention: the word segmentation processing of the text analysis in the data analysis unit specifically uses the Jieba word segmentation tool, and its word segmentation parameters and configuration are optimized and set according to the text characteristics and analysis requirements.

[0024] As a further solution of the present invention: the part-of-speech tagging of the text analysis in the data analysis unit specifically utilizes the NLTK library, and the part-of-speech tagging model and related resources in the library are selected and configured according to the specific language environment and analysis task.

[0025] As a further solution of the present invention: the named entity recognition of the text analysis in the data analysis unit adopts a method based on conditional random fields, and the training data, feature templates and parameter settings used in its training model are optimized according to the specific field and text type.

[0026] As a further solution of the present invention: a support vector machine classification model is constructed in the emotional tendency judgment, and the specific kernel function selection, parameter adjustment and training algorithm are optimized according to the task requirements and data characteristics of the emotional analysis.

[0027] As a further solution of the present invention: in the user behavior analysis, the association relationship between user behaviors is mined by an association rule mining algorithm, and the specific minimum support and minimum confidence thresholds are reasonably set according to the characteristics of the user behavior data and the analysis objectives to mine frequent item sets and strong association rules;

[0028] Among them, when setting the minimum support and minimum confidence thresholds, they are determined based on the characteristics of the user group being served and the data scale, using a method based on a combination of data distribution statistics and empirical rules. The mined frequent item sets and strong association rules will be used for user behavior pattern analysis.

[0029] As a further solution of the present invention: the user portrait is constructed based on the user's basic information and behavioral characteristics, and the specific basic information classification method, behavioral characteristic quantification method and feature vector construction rules are designed and determined according to actual application scenarios and analysis requirements.

[0030] As a further solution of the present invention: in the time series analysis, the changing trend of the user behavior data over time is analyzed, and the specific time series model selection, parameter estimation method and prediction algorithm are optimized according to the characteristics of the user behavior data and the prediction target.

[0031] As a further solution of the present invention: the node labels and relationship types created in the knowledge graph construction unit are defined and designed specifically according to the results of data analysis and domain knowledge, and in the process of inserting entities and association relationships into the graph database and creating indexes, the specific data format conversion, insertion operation and index creation strategy are optimized according to the characteristics and requirements of the graph database;

[0032] When building a knowledge graph:

[0033] First, according to the characteristics of multi-source data and the application requirements of the system, a reasonable node label system and relationship type definition are designed. Node labels include but are not limited to categories such as people, places, events, and products, and relationship types include belonging relationships, associated relationships, and time relationships.

[0034] Then, according to the customized data format requirements, the entity and relationship data obtained by named entity recognition and association relationship mining are cleaned and converted to ensure that the knowledge graph can accurately represent entities and relationships.

[0035] As a further solution of the present invention: the intelligent interaction unit converts user questions into semantic representations through semantic parsing technology, and specific semantic parsing algorithms, rules and models are selected and optimized according to the technical requirements and application scenarios of natural language processing.

[0036] As a further solution of the present invention: a convolutional neural network model is used in the intelligent interaction unit to classify the intent of user questions, and the specific network structure design, parameter setting and training strategy are optimized according to the task requirements and data characteristics of question classification.

[0037] As a further solution of the present invention: the personalized service unit adopts a collaborative filtering algorithm for personalized recommendation, and the specific user similarity calculation method, similar user group determination strategy and recommendation content screening rules are optimized according to user behavior data and recommendation goals.

[0038] The present invention also provides a method for implementing an intelligent assistant system, which is implemented by an intelligent assistant system and includes the following steps:

[0039] Step 1: Data Collection

[0040] Start the web crawler program, crawl relevant web page data according to the predetermined rules and frequency, and store it in the local database;

[0041] Regularly call third-party APIs to obtain required external data, and perform data integration and storage;

[0042] Record user behavior data in real time in the smart assistant application and upload it to the server for processing;

[0043] Step 2: Data Analysis

[0044] First, the collected data is cleaned;

[0045] Natural language processing techniques are then used to perform text analysis, including word segmentation, part-of-speech tagging, named entity recognition, sentiment analysis, and topic modeling.

[0046] Then conduct user behavior analysis to mine behavior patterns, build user portraits, and analyze behavior trends;

[0047] Step 3: Knowledge Graph Construction

[0048] First, based on the data analysis results, entities and relationships are extracted;

[0049] Then use the graph database to build a knowledge graph, and perform data storage and index creation;

[0050] Step 4: Intelligent Interaction

[0051] Semantic parsing and intent recognition realize the natural language understanding function, and then generate intelligent answers based on the text questions entered by the user and the knowledge graph, and support multiple rounds of dialogue;

[0052] Step 5: Personalized service

[0053] Personalized recommendations are achieved based on user portraits and behavior analysis.

[0054] Beneficial effects of the present invention:

[0055] The present invention collects multi-source data from multiple data sources, including various web page data obtained by web crawlers, covering news websites, forums, blogs, etc., and can obtain real-time news information and user comment text data to ensure the richness and timeliness of information. Access to third-party APIs can obtain data in professional fields, such as temperature, weather conditions, air quality and other weather information, to provide users with a wider range of service content. Collecting user behavior data during the use of smart assistants, including search keywords, click content, browsing time, etc., helps to gain a deeper understanding of user needs and behavioral habits.

[0056] The present invention cleans the collected text data, removes meaningless contents such as HTML tags, advertising information, special characters, etc., filters and replaces them through regular expressions or string processing functions, retains only meaningful text contents, improves data quality, and provides a reliable data basis for subsequent analysis.

[0057] The present invention adopts natural language processing technology to perform word segmentation, part-of-speech tagging and named entity recognition, and uses the Jieba word segmentation tool, NLTK library and conditional random field-based method to improve the accuracy and depth of text analysis. By building a support vector machine classification model to judge the emotional tendency of the text, it is helpful to understand the user's emotional attitude towards specific content and provide a basis for personalized services.

[0058] The present invention mines the correlation between user behaviors through association rule mining algorithm, builds user portraits based on users' basic information and behavioral characteristics, uses time series analysis method to analyze the changing trend of user behavior data over time, comprehensively understands user behavior patterns, and provides strong support for personalized recommendations.

[0059] The present invention screens and determines entities in a knowledge graph based on the results of named entity recognition in a data analysis unit, creates node labels and relationship types, imports entities and association relationships into a graph database, establishes a graph structure of the knowledge graph and creates an index, thereby improving the organization and retrieval efficiency of knowledge.

[0060] The present invention performs text parsing on natural language text input by the user, converts the user's question into a semantic representation understandable by a computer through semantic parsing technology, adopts a convolutional neural network model to classify the user's question and judge the user's intention, generates answers based on the knowledge graph and data analysis results, and improves the accuracy and intelligence of the answers; in the process of interaction with the user, it records the previous conversation content and context information, and answers according to the context, providing a more coherent and personalized interactive experience.

[0061] The present invention makes personalized recommendations based on user portraits and behavior analysis results, including recommendations based on user interests in specific types of content and in combination with user location information and time information, thereby improving the pertinence and practicality of recommendations; and adopts a collaborative filtering algorithm to make recommendations based on user behavior similarities, thereby providing users with content that meets their interests and needs and improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be further described below in conjunction with the accompanying drawings.

[0063] Figure 1 It is a system block diagram of an intelligent assistant system of the present invention.

[0064] Figure 2 It is a flowchart of a method for implementing an intelligent assistant system of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Embodiment 1

[0067] See also Figure 1 As shown, the present invention is an intelligent assistant system, comprising:

[0068] A data collection unit, used for collecting multi-source data from multiple data sources;

[0069] Among them, multi-source data includes:

[0070] Obtain various web page data from web crawlers, including news websites, forums, blogs, etc., to obtain text data corresponding to real-time information, user comments, etc.;

[0071] For example, use Python's Scrapy framework to write a crawler program, define the list of websites to be crawled and page rules, regularly crawl web page content and store it in a local database or data storage platform;

[0072] In this embodiment, it sets specific crawling frequency, depth and breadth strategies based on customized web crawler technology to obtain text data from various web pages that meet specific topics or fields, and at the same time follows the crawling rules and laws and regulations of relevant websites to avoid illegal or excessive crawling;

[0073] By cooperating with third-party APIs with legal authorization and data sharing agreements, we can obtain accurate text data in specific professional fields in accordance with the data interface specifications and security authentication mechanisms agreed upon by both parties to ensure the legality, accuracy and timeliness of the data;

[0074] For example, weather data API, map data API, etc., take obtaining weather data as an example, by registering and calling the weather forecast API, sending a request in the request format specified in the API document, you can obtain detailed weather information including temperature, weather conditions, air quality, etc.

[0075] Collect user behavior data when using smart assistants, such as text data corresponding to search keywords, clicked content, browsing time, etc. By embedding data collection code in the smart assistant application, when the user performs an operation, the relevant behavior data is automatically recorded and uploaded to the server for storage and analysis;

[0076] A preprocessing unit, used for cleaning the collected text data;

[0077] The text data includes but is not limited to HTML tags, advertising information, and special characters;

[0078] The cleaning process is to filter and replace text data through any of regular expressions or string processing functions, retaining only meaningful text content;

[0079] Specifically, the preprocessing unit uses regular expression-based matching and replacement rules to remove specific HTML tags and their contents. For advertising information, the pre-trained keyword recognition model is combined with the string processing function to identify and delete text fragments containing advertising keywords. For special characters, the string processing function is used to perform unified replacement or deletion operations based on the character encoding range and predefined rules.

[0080] Among them, regular expressions and string processing functions are both existing technologies, so they are not described in detail;

[0081] A data analysis unit, used for performing multi-faceted analysis on the pre-processed multi-source data;

[0082] The multi-faceted analysis method is as follows:

[0083] Step 1. Text Analysis

[0084] Step 1.1, use natural language processing technology to perform word segmentation, part-of-speech tagging and named entity recognition;

[0085] In this implementation:

[0086] First, use the Jieba word segmentation tool to segment the text. Through word segmentation, the sentence is divided into individual words.

[0087] Then use the NLTK library to perform part-of-speech tagging to determine the part-of-speech of each word, which includes but is not limited to nouns, verbs, and adjectives;

[0088] Then, a method based on conditional random fields is used for named entity recognition, which uses a pre-trained model to identify entities corresponding to names of people, places, and organizations in the text;

[0089] Step 1.2, determine the sentiment tendency of the text, where the sentiment tendency includes positive, negative or neutral;

[0090] The emotional tendency is judged as follows:

[0091] Construct a support vector machine classification model for emotional tendency judgment;

[0092] Firstly, we collect text data with accurate sentiment annotations from multiple data sources as training sets, and use the feature selection method based on information gain to select feature subsets with important distinguishing ability for sentiment classification from a variety of text features such as bag-of-words features and TF-IDF features.

[0093] Then, the parameters of the support vector machine classification model, such as kernel function type, penalty coefficient, etc., are optimized through cross-validation and grid search methods;

[0094] The trained support vector machine classification model can accurately judge the sentiment tendency of new texts based on the features extracted, providing sentiment analysis support for subsequent intelligent interaction and personalized services;

[0095] Step 2: User Behavior Analysis

[0096] Step 2.1, mining the association relationship between user behaviors through association rule mining algorithm;

[0097] In this implementation:

[0098] In user behavior data, analyze whether users search for nearby hotels or transportation information after searching for attractions;

[0099] Set the minimum support and minimum confidence thresholds to mine frequent itemsets and strong association rules from user behavior data;

[0100] For example, if it is found that many users often search for "nearby hotels" after searching for "Forbidden City", such an association rule can be established.

[0101] Step 2.2: Build a user profile based on the user's basic information (such as age, gender, region, etc.) and behavioral characteristics (such as browsing content, purchase history, search history, etc.);

[0102] Basic information includes but is not limited to the user's age, gender, and region;

[0103] Behavioral characteristics include, but are not limited to, users’ browsing content, purchase records, and search history;

[0104] For example, users are divided into different age groups, genders are divided into male and female, and regions are divided into provinces or cities. Behavioral characteristics are classified and quantified. The corresponding classified and quantified behavioral characteristics include the frequency of browsing scenic spot content and the amount of money spent on purchasing travel products. Then, this information is combined to form a feature vector for each user as a representation of the user portrait.

[0105] Step 2.3, use time series analysis method to analyze the changing trend of user behavior data over time;

[0106] For example, the number of users searching for different types of tourist destinations is counted in chronological order, and then the moving average method is used to predict future search trends using the pre-trained prediction model.

[0107] If it is found that the number of users' travel searches for coastal cities is on the rise recently, travel information and recommended content related to coastal cities will be recommended to users.

[0108] In this embodiment, multi-source data collection is rich and comprehensive, covering web pages, third-party APIs and user behavior data. Data cleaning improves quality, multi-faceted analysis is in-depth, text analysis helps understand content and judge emotions, and user behavior analysis mines associations, builds portraits and predicts trends, providing strong support for subsequent services.

[0109] Embodiment 2

[0110] As the second embodiment of the present invention, when the present application is implemented, compared with the first embodiment, the technical solution of the present embodiment is different from that of the first embodiment only in that the present embodiment further includes:

[0111] The knowledge graph construction unit is used to screen and determine entities in the knowledge graph based on the results of named entity recognition in the data analysis unit;

[0112] The specific method is as follows:

[0113] StepA.1. When building a knowledge graph, first create node labels and relationship types;

[0114] Design a reasonable node label system and relationship type definition based on the characteristics of multi-source data and the application requirements of the system;

[0115] Node labels include but are not limited to categories such as people, places, events, and products, and relationship types include belonging relationships, association relationships, time relationships, etc.

[0116] Then, according to the customized data format requirements, the entity and relationship data obtained by named entity recognition and relationship mining are cleaned and converted, and then the entities and relationships are inserted into the graph database one by one according to the data format requirements to establish the graph structure of the knowledge graph;

[0117] At the same time, an index is created based on the name attribute of the entity;

[0118] Step A.2, import the sorted entities and relationships into the graph database through the data import tool, and use the graph database to store the constructed knowledge graph;

[0119] Among them, entities are regarded as nodes, and relationships are regarded as edges. The attributes of entities and relationships are stored in nodes and edges in the form of attribute values.

[0120] For example, for a scenic spot node, attributes such as the name of the scenic spot, the description of the scenic spot, and the rating of the scenic spot can be stored; for an association edge of "the scenic spot is located in a city", the specific location relationship information between the scenic spot and the city can be stored;

[0121] An intelligent interaction unit for performing text parsing on natural language text input by the user;

[0122] The text is parsed as follows:

[0123] Step B.1. Convert the user's question into a semantic representation that can be understood by the computer through semantic parsing technology;

[0124] It uses a semantic parsing model based on deep learning. By learning a large amount of natural language text, the model can perform lexical and syntactic analysis on the natural language text input by the user, and use the pre-built semantic knowledge base to convert the user's question into a structured semantic representation that can be understood by the computer.

[0125] For example, for the user's question "What interesting attractions are there in Beijing?", semantic analysis is used to identify "Beijing" as a location entity and "interesting attractions" as a query requirement, and the corresponding semantic query structure is constructed.

[0126] Step B.2: Use the convolutional neural network model to classify user questions and determine user intentions;

[0127] The method is:

[0128] First, a large number of questions with different intentions are collected as training sets, and the questions are vectorized. In this embodiment, the vectorized representation uses the word2vec tool to convert words into vectors;

[0129] The vectorized problem is then fed into the convolutional neural network model for training;

[0130] The convolutional neural network model classifies the intent of new user questions by learning the characteristic patterns of the questions;

[0131] For example, it can identify the intent of searching for attractions, hotels, transportation, etc.

[0132] Step B.3, Generate answers based on the knowledge graph and data analysis results;

[0133] For example, when a user asks about a city's attractions, the system searches for the city's related attraction nodes in the knowledge graph, obtains the name and introduction of the attraction, and generates a natural language answer based on the popularity and user reviews of the attraction in the data analysis unit.

[0134] During the interaction with the user, record the previous conversation content and context information, and answer according to the context;

[0135] For example, if a user first asks about the weather in a city and then asks about attractions in the city that are suitable for outdoor activities, the system will take into account the previous inquiry about the weather when answering and recommend attractions suitable for outdoor activities under current weather conditions. By maintaining the conversation state and context information, the system can provide a more coherent and intelligent interactive experience.

[0136] This embodiment constructs a knowledge graph and filters entities based on named entity recognition results, which facilitates knowledge management and improves query efficiency. Intelligent interaction is enhanced by parsing natural language text, understanding the user's question intention, combining knowledge graphs and analysis results to generate answers, and recording context to provide a coherent interactive experience.

[0137] Embodiment 3

[0138] As the third embodiment of the present invention, when the present application is implemented, compared with the first and second embodiments, the technical solution of this embodiment is to combine the solutions of the first and second embodiments, and the difference between the technical solution of this embodiment and the first and second embodiments is that this embodiment also includes:

[0139] Personalized service unit, used to make personalized recommendations based on user portraits and behavior analysis results;

[0140] For example, if the user profile shows that the user is interested in historical and cultural attractions and has recently browsed a lot of content about Beijing's history and culture, the system will give priority to recommending Beijing's historical and cultural attractions, such as the Forbidden City and the Summer Palace;

[0141] At the same time, combined with the user's location information and time information, if the user is currently in Beijing and it is the weekend, the system can also recommend nearby related tourist activities or hotel information;

[0142] Personalized recommendations use collaborative filtering algorithms, which make recommendations based on user behavior similarities;

[0143] The specific method is as follows:

[0144] First, by analyzing the user's behavior data, extract the user's behavior characteristics, such as the user's browsing history, purchase history, search keywords, etc.;

[0145] Then, a similarity calculation method, such as cosine similarity or Pearson correlation coefficient, is used to calculate the behavioral similarity between the target user and other users, and to find a user group with similar behavior to the target user;

[0146] Based on the preferences and behavior frequencies of these similar users for different contents, the target user's interest in the unexposed content is predicted by weighted average or other methods;

[0147] Finally, recommend content that matches the target user's potential interests in descending order of interest;

[0148] For example, the cosine similarity algorithm can be used to calculate the similarity between users. User feature vectors are constructed based on the user's ratings or browsing behaviors for different attractions, hotels, etc., and then the cosine similarity between the vectors is calculated;

[0149] Recommend content that is liked by user groups with high similarity but has not been exposed to the target users.

[0150] This embodiment makes personalized recommendations based on user portraits and behavior analysis, combining interests, location and time information to accurately recommend content. A collaborative filtering algorithm is used to recommend based on behavior similarity to improve recommendation accuracy and personalization.

[0151] Embodiment 4

[0152] As the fourth embodiment of the present invention, when the present application is specifically implemented, compared with the first, second and third embodiments, the technical solution of this embodiment is to combine and implement the solutions of the above-mentioned first, second and third embodiments.

[0153] This embodiment combines the advantages of the previous three embodiments to achieve comprehensive data collection, in-depth analysis, knowledge graph construction, intelligent interaction and personalized services, providing users with a more comprehensive, intelligent and personalized service experience.

[0154] See also Figure 2 As shown, the present invention also provides a method for implementing an intelligent assistant system, which is implemented by an intelligent assistant system and includes the following steps:

[0155] Step 1: Data Collection

[0156] Start the web crawler program, crawl relevant web page data according to the predetermined rules and frequency, and store it in the local database;

[0157] Regularly call third-party APIs to obtain required external data, and perform data integration and storage;

[0158] Record user behavior data in real time in the smart assistant application and upload it to the server for processing;

[0159] Step 2: Data Analysis

[0160] First, the collected data is cleaned;

[0161] Natural language processing techniques are then used to perform text analysis, including word segmentation, part-of-speech tagging, named entity recognition, sentiment analysis, and topic modeling.

[0162] Then conduct user behavior analysis to mine behavior patterns, build user portraits, and analyze behavior trends;

[0163] Step 3: Knowledge Graph Construction

[0164] First, based on the data analysis results, entities and relationships are extracted;

[0165] Then use the graph database to build a knowledge graph, and perform data storage and index creation;

[0166] Step 4: Intelligent Interaction

[0167] Semantic parsing and intent recognition realize the natural language understanding function, and then generate intelligent answers based on the text questions entered by the user and the knowledge graph, and support multiple rounds of dialogue;

[0168] Step 5: Personalized service

[0169] Personalized recommendations are achieved based on user portraits and behavior analysis.

[0170] In the above content, the related technical features that are not further described are all prior arts, so they will not be described in detail; and the above description is only a specific implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An intelligent assistant system, characterized in that: include: A data collection unit, used to collect multi-source data, where the multi-source data is text data collected from multiple data sources; A preprocessing unit, used for cleaning the collected text data; The data analysis unit is used to conduct multi-faceted analysis on the pre-processed multi-source data. It uses natural language processing technology to perform word segmentation, part-of-speech tagging and named entity recognition, and then judges the sentiment tendency of the text; at the same time, it uses association rule mining algorithms to mine the correlation between user behaviors; and also builds user portraits based on the user's basic information and behavioral characteristics; And use time series analysis methods to analyze the changing trends of user behavior data over time; The knowledge graph construction unit is used to screen and determine entities in the knowledge graph based on the results of named entity recognition in the data analysis unit; The intelligent interaction unit is used to perform text analysis on the natural language text input by the user. The method is to use a semantic analysis model based on deep learning. The model performs lexical and syntactic analysis on the natural language text input by the user through learning the natural language text, and uses the pre-built semantic knowledge base to convert the user's question into a structured semantic representation that can be understood by the computer. Then, a convolutional neural network model is used to classify the user's question and determine the user's intention, and then an answer is generated based on the knowledge graph and data analysis results. Personalized service unit, used to make personalized recommendations based on user portraits and behavior analysis results; Personalized recommendations use collaborative filtering algorithms, which make recommendations based on the similarity of user behaviors.

2. The intelligent assistant system according to claim 1, characterized in that: in, Multi-source data includes: Obtain text data corresponding to various web pages based on web crawler technology; Access third-party APIs to obtain text data corresponding to professional fields; Collect user behavior data when using smart assistants. The behavior data includes text data corresponding to the user's search keywords, the content the user clicks, and the user's browsing time.

3. The intelligent assistant system according to claim 2, characterized in that: in, Text data includes but is not limited to HTML tags, advertising information, and special characters; The cleaning treatment method of the pretreatment unit is: For HTML tags, regular expression-based matching and replacement rules are used to remove specific HTML tags and their contents; For advertising information, a pre-trained keyword recognition model is combined with a string processing function to identify and delete text segments containing advertising keywords. For special characters, string processing functions are used to perform unified replacement or deletion operations based on the character encoding range and predefined rules.

4. The intelligent assistant system according to claim 1, characterized in that: In the multifaceted analysis of the data analysis unit, word segmentation is performed on the text by using the Jieba word segmentation tool; Part-of-speech tagging uses the NLTK library to determine the part-of-speech of each word after segmentation. The parts of speech include but are not limited to nouns, verbs, and adjectives. Named entity recognition is performed using a conditional random field-based method, which uses a pre-trained model to identify entities corresponding to names of people, places, and organizations in the text.

5. The intelligent assistant system according to claim 1, characterized in that: in, Sentimental tendencies include positive, negative, or neutral. Sentimental tendencies are determined as follows: Build a support vector machine classification model; First, a large amount of text data with sentiment annotations is collected as a training set, and text features are extracted. The text features include but are not limited to bag-of-words features and TF-IDF features. Then, a support vector machine classification model is trained, and the support vector machine classification model is used to determine the sentiment category based on text features; Similarly, the text features corresponding to the new text are extracted and input into the pre-trained support vector machine classification model for sentiment prediction to determine the sentiment tendency of the new text.

6. The intelligent assistant system according to claim 1, characterized in that: The mining method of association relationship is as follows: In user behavior data, analyze whether users search for information related to a piece of content after searching for that content; At the same time, the minimum support and minimum confidence thresholds are set to mine frequent item sets and strong association rules from user behavior data, and the mined frequent item sets and strong association rules will be used for user behavior pattern analysis; Among them, when setting the minimum support and minimum confidence thresholds, they are determined based on the characteristics of the user group being served and the scale of data, using a method based on a combination of data distribution statistics and empirical rules.

7. The intelligent assistant system according to claim 1, characterized in that: The method of constructing the knowledge graph unit is as follows: When building a knowledge graph, first create node labels and relationship types; Then, according to the data format requirements, the entities and relationships are inserted into the graph database one by one to establish the graph structure of the knowledge graph; At the same time, an index is created based on the name attribute of the entity; Then, the sorted entities and relationships are imported into the graph database through the data import tool, and the constructed knowledge graph is stored in the graph database; In this method, entities are regarded as nodes, relationships are regarded as edges, and the attributes of entities and relationships are stored in nodes and edges in the form of attribute values.

8. The intelligent assistant system according to claim 1, characterized in that: The user intention is determined by: First, a large number of questions with different intentions are collected as training sets, and the questions are vectorized. The vectorized problem is then fed into the convolutional neural network model for training; The convolutional neural network model classifies the intent of new user questions by learning the characteristic patterns of the questions.

9. The intelligent assistant system according to claim 1, characterized in that: The specific methods of personalized recommendation are as follows: Through collaborative filtering algorithms, we find other user groups with similar behaviors to the target user, and then recommend content that may be of interest to the target user based on the preferences and behaviors of these similar users.

10. A method for implementing an intelligent assistant system, characterized in that: The method is implemented by the intelligent assistant system described in any one of claims 1 to 9.

Images