An AI-based intelligent project consulting system and method

Abstract

This invention relates to the field of project recommendation technology, specifically to an artificial intelligence-based intelligent project consultation system and method. The system includes an information collection unit, a consultation matching unit, a project scoring unit, a feature extraction unit, a project classification unit, a project recommendation unit, and a scoring update unit. Specifically, the information collection unit acquires project consultation information from a target user based on a preset project consultation template and transmits this information to the consultation matching unit. This invention obtains the user's specific consultation needs through the information collection unit and matches them with a preset project database through the consultation matching unit. The system can provide a set of candidate projects most relevant to the user's needs. Then, through historical scoring information and scoring feature extraction from the project scoring unit and feature extraction unit, the recommended projects become more personalized and accurate, improving the user experience.

Description

Technical Field

[0001] This invention relates to the field of project recommendation technology, specifically to an intelligent project consulting system and method based on artificial intelligence. Background Technology

[0002] Artificial intelligence (AI) is a branch of computer science that aims to enable machines to perform tasks that typically require human intelligence, including learning, reasoning, problem-solving, language understanding, visual recognition, and decision-making. Simply put, the goal of AI is to enable machines to think, understand, and act like humans, and even surpass human capabilities in some aspects.

[0003] Traditional systems typically rely on simple rules or keyword-based matching. These methods are usually static and lack a deep understanding of users' personalized needs, resulting in inaccurate recommendations that fail to fully meet specific user requirements. Furthermore, traditional systems depend on user ratings and reviews for item recommendations, but these ratings are often influenced by fake ratings, malicious reviews, or distorted information. Without an effective mechanism to distinguish between genuine and fake ratings, recommendation systems are easily affected by this inaccurate information. Moreover, traditional systems often rely on fixed rating models or simple rules, lacking dynamic adjustment capabilities. Rating updates are usually done manually, requiring human intervention, and are infrequent, causing the system to fail to respond promptly to changes in market and user behavior. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide an intelligent project consulting system and method based on artificial intelligence.

[0005] The technical solution adopted to solve the above-mentioned technical problems is: an intelligent project consulting system based on artificial intelligence, including an information collection unit, a consulting matching unit, a project scoring unit, a feature extraction unit, a project classification unit, a project recommendation unit, and a scoring update unit, specifically:

[0006] The information collection unit is used to obtain the project consultation information of the target consultation user based on the preset project consultation template, and transmit the project consultation information of the target consultation user to the consultation matching unit;

[0007] The consultation matching unit is used to receive the project consultation information of the target consultation user transmitted by the information collection unit, match the project consultation information of the target consultation user with a preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user, and transmit the set of candidate projects to the project scoring unit.

[0008] The project scoring unit is used to receive the candidate project set transmitted by the consultation matching unit, and to obtain the historical scoring information of each candidate project in the candidate project set, wherein the historical scoring information includes historical scores and scoring users, and to transmit the historical scoring information of each candidate project in the candidate project set to the feature extraction unit.

[0009] Preferably, the feature extraction unit is used to receive the historical rating information of each candidate item in the candidate item set transmitted by the project rating unit, and to extract features from the historical rating information of each candidate item in the candidate item set to obtain a rating feature set of each candidate item in the candidate item set, and to transmit the rating feature set of each candidate item in the candidate item set to the project classification unit.

[0010] Preferably, the project classification unit receives the rating feature set of each candidate project in the candidate project set transmitted by the feature extraction unit, and classifies the rating feature set of each candidate project in the candidate project set based on a pre-trained rating classification model to obtain the rating type label corresponding to each candidate project in the candidate project set. The rating type label includes a real rating label and a fake rating label. Based on the project type label, the candidate project set is divided into a real rating project set and a fake rating project set. The real rating project set is transmitted to the project recommendation unit, and the real rating project set and the fake rating project set are transmitted to the rating update unit.

[0011] Preferably, the project recommendation unit is used to receive the set of real-rated projects transmitted by the project classification unit, and to recommend projects to the target consulting user based on the set of real-rated projects, so as to obtain a project recommendation list corresponding to the target consulting user.

[0012] Preferably, the rating update unit is used to receive the set of real rating items and the set of fake rating items transmitted by the item classification unit, and match the set of fake rating items and the set of real rating items based on a similarity model to obtain a real rating item that matches each fake rating item in the set of fake rating items, and update the rating of each fake rating item in the set of fake rating items based on the rating of the real rating item.

[0013] Preferably, feature extraction is performed on the historical rating information of each candidate item in the candidate item set to obtain a rating feature set for each candidate item in the candidate item set, including:

[0014] Calculate the project popularity of each candidate project in the candidate project set based on the historical rating information;

[0015] Calculate the project novelty of each candidate project in the candidate project set based on the historical scoring information;

[0016] The candidate project set is sorted according to the project popularity, and the candidate project set is divided into a popularity set and a non-popular set based on cross-validation.

[0017] The candidate project set is sorted according to the novelty of the project, and the candidate project set is divided into a novel set and a non-novel set based on cross-validation;

[0018] Calculate the popularity distribution of each popularity item in the popularity set based on the historical rating information;

[0019] The non-hotness distribution of each non-hotness item in the non-hotness set is calculated based on the historical rating information.

[0020] Calculate the novelty distribution of each novel item in the novelty set based on the historical scoring information;

[0021] Calculate the non-novel distribution of each non-novel item in the non-novel set based on the historical scoring information;

[0022] The popularity distribution, the non-popularity distribution, the novelty distribution, and the non-popularity distribution are combined to obtain the scoring feature set for each candidate item in the candidate item set.

[0023] Preferably, the formula for calculating the project's popularity is as follows:

[0024] ;

[0025] in, This indicates the ratings of all users for the candidate items. Number of ratings Indicates rating user For candidate projects The rating value, when ,but Otherwise, , Represents the set of all users who rate the service;

[0026] The formula for calculating the novelty of the project is as follows:

[0027] ;

[0028] in, This represents all user rating candidate items. Dissimilarity to other different rating candidate items express The total number of projects in China Indicates rating user The number of candidate items for scoring Indicates candidate projects With candidate projects The similarity between them;

[0029] The formula for calculating the heat distribution is as follows:

[0030] ;

[0031] in, Indicates candidate items in the heat set The proportion of popularity to the total popularity of the entire popularity set, and , Represents a set of heat values;

[0032] The formula for calculating the non-thermal distribution is as follows:

[0033] ;

[0034] in, Indicates candidate items in the non-hot set The proportion of popularity of non-popularity aggregates to the total popularity of [the data / intensity]. This represents a set of non-heat sets.

[0035] Preferably, the novel distribution is calculated using the following formula:

[0036] ;

[0037] in, Represents candidate items in the novel set The proportion of the novelty of the novelty to the total novelty of the novelty set, and , Represents a novel set;

[0038] The formula for calculating the non-novel distribution is as follows:

[0039] ;

[0040] in, Indicates candidate items in a non-novel set The proportion of novelty to the total novelty of the non-novel set. This represents a non-novel set.

[0041] Preferably, the rating classification model employs an improved support vector machine (SVM) model. The penalty parameters and kernel function parameters of the SVM model are optimized using the Grey Wolf optimization algorithm to obtain the rating classification model. Specifically, optimizing the penalty parameters and kernel function parameters of the SVM model using the Grey Wolf optimization algorithm to obtain the rating classification model includes:

[0042] Randomly initialize individuals in the gray wolf population, where the position of each individual represents the parameter combination of the penalty parameter and kernel function parameter of the support vector machine model;

[0043] The positions of the individual gray wolves in the population are updated to obtain new positions for the individual gray wolves. The formula for updating the new positions is as follows:

[0044] ;

[0045] in, Indicates the Position at the next iteration Indicates the location of the prey. Denotes the first random parameter, and , Represents the nonlinear convergence factor. express The first random number within the interval, Indicates the distance between the prey and the wolf pack, and express, Indicates the Position at the next iteration Denotes the second random parameter, and , express The second random number within the interval;

[0046] Calculate the new fitness of the gray wolf population individual at the new location and compare it with the optimal fitness of the previous iteration. If the new fitness is greater than the optimal fitness, replace the new fitness with the optimal fitness and retain the new location of the gray wolf population individual.

[0047] If the number of iterations exceeds the preset iteration threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters; otherwise, the above operation is repeated.

[0048] Preferably, based on the set of real-rated items, item recommendations are made to the target consulting user to obtain an item recommendation list corresponding to the target consulting user, including:

[0049] Obtain the set of rating users corresponding to the set of real rating items;

[0050] Calculate the user similarity between the target consulting user and the set of rating users;

[0051] Based on the user similarity, obtain the K-nearest neighbor set of the target consulting user;

[0052] Calculate the true rating item deviation based on the true ratings corresponding to the K nearest neighbor set of the target consulting users;

[0053] Calculate the predicted rating of the target consulting user for the actual rating item based on the deviation of the actual rating item;

[0054] The set of real-rated items is sorted according to the predicted ratings, and the top preset number of real-rated items in the set of real-rated items are selected as recommended items to obtain the recommended list of items corresponding to the target consulting user.

[0055] Preferably, the formula for calculating user similarity is as follows:

[0056] ;

[0057] in, Indicates rating user and target consulting users In classification Similarity on Indicates rating user and target consulting users Those who were also evaluated belong to the category. Project collection, Indicates user Belonging to The average rating of the items in the list, Indicates target consulting users Belonging to The average score of the items in the list;

[0058] The formula for calculating the deviation of the actual rating items is as follows:

[0059] ;

[0060] in, Indicates belonging to the same category Project and projects The average deviation, Indicates that they belong to the same category Project and projects The set of users who rate simultaneously Indicates belonging to the same category Below and target consulting users The set of the K most similar nearest neighbors;

[0061] The formula for calculating the predicted score is as follows:

[0062] ;

[0063] in, Indicates target consulting users For the project Predicted score Indicates excluding projects In addition, target consulting users Rated categories A collection of projects.

[0064] The technical solution adopted to solve the above-mentioned technical problems is: an artificial intelligence-based intelligent project consulting method, which is applicable to the aforementioned artificial intelligence-based intelligent project consulting system, including:

[0065] Based on a preset project consultation template, the project consultation information of the target consultation user is obtained, and the project consultation information of the target consultation user is matched with a preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user.

[0066] Obtain historical rating information for each candidate item in the candidate item set, wherein the historical rating information includes historical ratings and rating users;

[0067] Feature extraction is performed on the historical rating information of each candidate item in the candidate item set to obtain the rating feature set of each candidate item in the candidate item set;

[0068] The rating feature set of each candidate item in the candidate item set is classified based on a pre-trained rating classification model to obtain the rating type label corresponding to each candidate item in the candidate item set. The rating type label includes a real rating label and a fake rating label. Based on the item type label, the candidate item set is divided into a real rating item set and a fake rating item set.

[0069] Based on the set of real-rated items, project recommendations are made to the target consulting user to obtain a project recommendation list corresponding to the target consulting user;

[0070] The set of fake rating items and the set of real rating items are matched based on a similarity model to obtain real rating items that match each fake rating item in the set of fake rating items. The rating of each fake rating item in the set of fake rating items is updated based on the rating of the real rating items.

[0071] The beneficial effects of the present invention are as follows: (1) The present invention obtains the user's specific consultation needs through the information collection unit and matches them with the preset project library through the consultation matching unit. The system can provide the most relevant candidate project set according to the user's needs. Then, through the historical rating information and rating feature extraction of the project rating unit and feature extraction unit, the recommended projects are more personalized and accurate, thus improving the user experience; (2) The present invention classifies the rating information of the candidate projects through the project classification unit, identifies false rating projects, and removes them from the recommendation list. This classification method can effectively reduce the interference of false ratings or distorted information on the recommendation results, and ensure that the items recommended to the user are accurate. The purpose is to improve the reliability and trustworthiness of the recommendation system by using a rating update unit to match false ratings and real ratings based on a similarity model and update the ratings of false ratings. This can effectively and dynamically adjust the rating system and further improve the accuracy of the ratings. This allows the system to automatically optimize and update the rating model at any time based on newly collected information, maintaining the real-time performance and accuracy of the system. Moreover, it adopts a multi-level automated processing flow, reducing the need for manual intervention and improving efficiency and processing speed. Through artificial intelligence algorithms, the system can continuously learn and improve, making each recommendation more in line with the user's needs. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention;

[0073] Figure 2 This is a schematic diagram of the overall method steps in one embodiment of the present invention.

[0074] Attached labels: 1. Information collection unit; 2. Consultation matching unit; 3. Project scoring unit; 4. Feature extraction unit; 5. Project classification unit; 6. Project recommendation unit; 7. Scoring update unit. Detailed Implementation

[0075] Example 1, as Figure 1 As shown, the present invention proposes an intelligent project consulting system based on artificial intelligence, comprising an information collection unit 1, a consulting matching unit 2, a project scoring unit 3, a feature extraction unit 4, a project classification unit 5, a project recommendation unit 6, and a scoring update unit 7, specifically:

[0076] Information collection unit 1 is used to obtain project consultation information of target consultation users based on preset project consultation templates, and transmit the project consultation information of target consultation users to consultation matching unit 2;

[0077] Consultation matching unit 2 is used to receive the project consultation information of the target consultation user transmitted by information collection unit 1, match the project consultation information of the target consultation user with the preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user, and transmit the set of candidate projects to project scoring unit 3.

[0078] The project scoring unit 3 is used to receive the candidate project set transmitted by the consultation matching unit 2 and obtain the historical scoring information of each candidate project in the candidate project set. The historical scoring information includes historical scores and scoring users. The historical scoring information of each candidate project in the candidate project set is transmitted to the feature extraction unit 4.

[0079] In this invention, the target consulting user refers to those seeking project advice, consultation, or matching, typically a business or individual; project consulting information refers to the user's submitted needs, conditions, or expectations regarding a project, including project type, budget, objectives, timeframe, resource requirements, etc.; the project library is a collection containing multiple project data, typically including project name, description, budget, timeframe, historical execution status, etc., and is usually a database in the system for storing and managing projects; the candidate project set is a set of projects selected by a matching algorithm, which meet the needs of the target consulting user or meet the matching criteria; in short, these projects are potential candidates for the user's needs; historical rating information refers to the scores given by rating users when rating a project in the past.

[0080] In an optional embodiment, the feature extraction unit 4 is used to receive the historical rating information of each candidate item in the candidate item set transmitted by the project rating unit 3, and to extract features from the historical rating information of each candidate item in the candidate item set to obtain the rating feature set of each candidate item in the candidate item set, and to transmit the rating feature set of each candidate item in the candidate item set to the project classification unit 5.

[0081] It should be noted that the rating feature set refers to the set of all rating-related features extracted from historical rating information by the feature extraction unit.

[0082] In an optional embodiment, the project classification unit 5 receives the rating feature set of each candidate project in the candidate project set transmitted by the feature extraction unit 4, and classifies the rating feature set of each candidate project in the candidate project set based on a pre-trained rating classification model to obtain the rating type label corresponding to each candidate project in the candidate project set. The rating type label includes a real rating label and a fake rating label. Based on the project type label, the candidate project set is divided into a real rating project set and a fake rating project set. The real rating project set is transmitted to the project recommendation unit 6, and the real rating project set and the fake rating project set are transmitted to the rating update unit 7.

[0083] It's important to note that a rating classification model is a machine learning model designed to categorize ratings based on their characteristics. This model is typically trained by analyzing large amounts of historical rating data and learns to distinguish between different types of ratings. The goal of the classification model is to predict whether a rating belongs to a specific category, such as "real rating" or "fake rating." Rating type labels refer to the labels used to categorize a project's ratings. Based on the output of the rating classification model, there are two types of rating type labels: real rating labels, which indicate that the project's ratings come from real users and reflect the project's true quality; and fake rating labels, which indicate that the project's ratings may originate from fake or manipulated rating behavior.

[0084] In an optional embodiment, the project recommendation unit 6 is used to receive the set of real-rated projects transmitted by the project classification unit 5, and to recommend projects to the target consulting user based on the set of real-rated projects, so as to obtain a project recommendation list corresponding to the target consulting user.

[0085] In an optional embodiment, the rating update unit 7 is used to receive the set of real rating items and the set of fake rating items transmitted by the item classification unit 5, and match the set of fake rating items and the set of real rating items based on a similarity model to obtain the real rating items that match each fake rating item in the set of fake rating items, and update the rating of each fake rating item in the set of fake rating items based on the rating of the real rating items.

[0086] It's important to note that a similarity model is an algorithm used to measure the similarity between two objects. It determines their similarity by comparing their features. Similarity models are commonly used in recommendation systems to discover similarities between users or items. Matching refers to pairing or associating fake rated items with real rated items. In this problem, the goal of matching is to find real rated items similar to the fake rated items. Rating update refers to adjusting the rating of an item according to certain rules or algorithms. The rating of each item in the set of fake rated items is updated based on the ratings of real rated items that are similar to it. Suppose that a fake rated item was manipulated during the rating process, its rating may not be accurate. By comparing it with the matched real rated items, the rating of the fake rated item can be updated to better align with the rating trends of real items, thereby improving the credibility of the rating.

[0087] Example 2: The present invention proposes an intelligent project consulting system based on artificial intelligence. Compared with Example 1, this example further includes: extracting features from the historical rating information of each candidate project in the candidate project set to obtain a rating feature set for each candidate project in the candidate project set, including:

[0088] The project popularity of each candidate project in the candidate project set is calculated based on historical rating information.

[0089] The project novelty of each candidate project in the candidate project set is calculated based on historical scoring information;

[0090] The candidate project set is sorted according to the project popularity, and the candidate project set is divided into a popular set and a non-popular set based on cross-validation;

[0091] The candidate project set is sorted according to the novelty of the projects, and the candidate project set is divided into a novel set and a non-novel set based on cross-validation;

[0092] Calculate the popularity distribution of each popular item in the popularity set based on historical rating information;

[0093] The non-hotness distribution of each non-hotness item in the non-hotness set is calculated based on historical rating information;

[0094] Calculate the novelty distribution of each novel item in the novelty set based on historical rating information;

[0095] Calculate the non-novel distribution of each non-novel item in the non-novel set based on historical rating information;

[0096] The popularity distribution, non-popularity distribution, novelty distribution, and non-popularity distribution are combined to obtain the score feature set for each candidate item in the candidate item set.

[0097] In this embodiment, project popularity is an indicator that measures the popularity of a project in historical rating data. Generally speaking, projects with high popularity are those that have been rated or visited by a large number of users, which may represent the current mainstream user preferences. Project novelty measures the "freshness" of a project to users in the recommendation list, that is, whether the project is something that users have not encountered, is unique, or has few ratings. Projects with high novelty are usually newly released, niche, or projects that users have never encountered before. Popularity set refers to the set of projects considered popular and with a large number of ratings in a candidate project set after sorting according to historical rating information. Usually, these projects have high historical ratings and may have received widespread attention and interaction. Non-popularity set refers to those projects with low popularity and a small number of ratings. A smaller set of projects typically consists of less popular, less mainstream projects; a novel set refers to the set of projects with higher novelty levels after ranking them by their novelty. A non-novel set refers to the set of projects with lower novelty levels after ranking them by their novelty levels. Popularity distribution refers to the distribution of a project's ratings within the popular set in historical rating information. Non-popularity distribution refers to the distribution of a project's ratings within the non-popular set in historical rating information. Novelty distribution refers to the distribution of a project's ratings within the novel set in historical rating information. Non-novelty distribution refers to the distribution of a project's ratings within the non-novel set in historical rating information.

[0098] In an optional embodiment, the formula for calculating project popularity is as follows:

[0099] ;

[0100] in, This indicates the ratings of all users for the candidate items. Number of ratings Indicates rating user For candidate projects The rating value, when ,but Otherwise, , Represents the set of all users who rate the service;

[0101] The formula for calculating the novelty of a project is as follows:

[0102] ;

[0103] in, This represents all user rating candidate items. Dissimilarity to other different rating candidate items express The total number of projects in China Indicates rating user The number of candidate items for scoring Indicates candidate projects With candidate projects The similarity between them;

[0104] The formula for calculating heat distribution is as follows:

[0105] ;

[0106] in, Indicates candidate items in the heat set The proportion of popularity to the total popularity of the entire popularity set, and , Represents a set of heat values;

[0107] The formula for calculating non-thermal distribution is as follows:

[0108] ;

[0109] in, Indicates candidate items in the non-hot set The proportion of popularity of non-popularity aggregates to the total popularity of [the data / intensity]. This represents a set of non-heat sets.

[0110] In an optional embodiment, the novel distribution is calculated as follows:

[0111] ;

[0112] in, Represents candidate items in the novel set The proportion of the novelty of the novelty to the total novelty of the novelty set, and , Represents a novel set;

[0113] The formula for calculating the non-novel distribution is as follows:

[0114] ;

[0115] in, Indicates candidate items in a non-novel set The proportion of novelty to the total novelty of the non-novel set. This represents a non-novel set.

[0116] In an optional embodiment, the rating classification model employs an improved support vector machine (SVM) model. The penalty parameters and kernel function parameters of the SVM model are optimized using the Grey Wolf optimization algorithm to obtain the rating classification model. The optimization of the penalty parameters and kernel function parameters of the SVM model using the Grey Wolf optimization algorithm to obtain the rating classification model includes:

[0117] Randomly initialize individuals in the gray wolf population. The position of each individual in the gray wolf population represents the parameter combination of the penalty parameter and kernel function parameter of the support vector machine model.

[0118] The positions of individual gray wolf individuals are updated to obtain their new positions. The formula for updating the new positions is as follows:

[0119] ;

[0120] in, Indicates the Position at the next iteration Indicates the location of the prey. Denotes the first random parameter, and , Represents the nonlinear convergence factor. express The first random number within the interval, Indicates the distance between the prey and the wolf pack, and express, Indicates the Position at the next iteration Denotes the second random parameter, and , express The second random number within the interval;

[0121] Calculate the new fitness of the gray wolf population at the new location and compare it with the best fitness of the previous iteration. If the new fitness is greater than the best fitness, replace the new fitness with the best fitness and retain the new location of the gray wolf population.

[0122] If the number of iterations exceeds the preset iteration threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters; otherwise, the above operation is repeated.

[0123] It's important to note that Support Vector Machines (SVMs) are supervised learning algorithms used for classification and regression tasks. They aim to classify data by maximizing the margin between classes (i.e., the decision boundary). SVMs use one or more kernel functions to map data into a high-dimensional space, making it linearly separable. The penalty parameter, also called the constant parameter, is a hyperparameter in SVM that controls the model's tolerance for misclassification. A larger C value makes the model fit the training data more tightly, potentially leading to overfitting, while a smaller C value may cause underfitting. Kernel functions are used to map data to higher dimensions to achieve linear segmentation in the high-dimensional space. Common kernel functions include radial basis function (RBF) kernels and polynomial kernels. The parameters of the kernel function (such as the γ parameter in the RBF kernel) affect the structure of the mapped high-dimensional space. The Grey Wolf Optimization (GWO) algorithm is a heuristic optimization algorithm that simulates the hunting behavior of grey wolves in packs. GWO searches for the optimal solution to a problem by simulating the hunting process among grey wolves (divided into leader wolves, follower wolves, and prey). This algorithm is often used to solve complex optimization problems. In optimization algorithms, fitness is a standard for measuring the quality of an individual (or solution). Usually, the fitness function is defined according to the objective of the problem and represents the quality of a solution. In gray wolf optimization, fitness reflects the quality of the individual solution of the gray wolf and is usually related to the objective function (e.g., classification accuracy).

[0124] In an optional embodiment, project recommendations are made to the target consulting user based on a set of real-rated projects to obtain a project recommendation list corresponding to the target consulting user, including:

[0125] Obtain the set of rating users corresponding to the set of real rating items;

[0126] Calculate user similarity between the target consultation users and the set of rating users;

[0127] Obtain the K-nearest neighbor set of the target consulting user based on user similarity;

[0128] Calculate the true rating item bias based on the true ratings corresponding to the K nearest neighbor set of the target consulting users;

[0129] Calculate the target consulting user's predicted rating for the actual rated items based on the deviation of the actual rated items;

[0130] The set of real-rated items is sorted according to the predicted ratings, and the top preset number of real-rated items in the set of real-rated items are selected as recommended items to obtain a recommended list of items corresponding to the target consulting user.

[0131] It's important to note that the K-Nearest Neighbors (KNN) algorithm is a commonly used similarity-based algorithm. The K-nearest neighbor set refers to the set of the K users with the highest similarity to the target user; users in this set are considered valuable for recommendations to the target user. Bias refers to the difference between a user's rating of a particular item and the ratings of other users in their K-nearest neighbor set for that item. Predicted rating refers to estimating the target user's likely rating for a particular item based on the behavioral data of the target user and other users in their K-nearest neighbor set. Ranking refers to sorting the predicted ratings, usually by their highest score, placing items most likely to be liked by the target user at the top. The recommended item list is a personalized set of recommended items generated based on the user's interests and predicted ratings.

[0132] In an optional embodiment, the formula for calculating user similarity is as follows:

[0133] ;

[0134] in, Indicates rating user and target consulting users In classification Similarity on Indicates rating user and target consulting users Those who were also evaluated belong to the category. Project collection, Indicates user Belonging to The average rating of the items in the list, Indicates target consulting users Belonging to The average score of the items in the list;

[0135] The formula for calculating the deviation of the actual rating items is as follows:

[0136] ;

[0137] in, Indicates belonging to the same category Project and projects The average deviation, Indicates that they belong to the same category Project and projects The set of users who rate simultaneously Indicates belonging to the same category Below and target consulting users The set of the K most similar nearest neighbors;

[0138] The formula for calculating the predicted score is as follows:

[0139] ;

[0140] in, Indicates target consulting users For the project Predicted score Indicates excluding projects In addition, target consulting users Rated categories A collection of projects.

[0141] Example 3, as Figure 2 As shown, the present invention proposes an artificial intelligence-based intelligent project consulting method, which is applicable to the aforementioned artificial intelligence-based intelligent project consulting system, comprising:

[0142] S1. Obtain the project consultation information of the target consultation user based on the preset project consultation template, and match the project consultation information of the target consultation user with the preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user.

[0143] S2. Obtain the historical rating information for each candidate item in the candidate item set, where the historical rating information includes historical ratings and rating users;

[0144] S3. Extract features from the historical rating information of each candidate item in the candidate item set to obtain the rating feature set of each candidate item in the candidate item set.

[0145] S4. Based on the pre-trained rating classification model, classify the rating feature set of each candidate item in the candidate item set to obtain the rating type label corresponding to each candidate item in the candidate item set. The rating type label includes real rating label and fake rating label. Based on the item type label, divide the candidate item set into a real rating item set and a fake rating item set.

[0146] S5. Based on the set of real-rated items, recommend items to the target consulting users to obtain a list of recommended items corresponding to the target consulting users;

[0147] S6. Match the set of fake rating items and the set of real rating items based on the similarity model to obtain the real rating items that match each fake rating item in the set of fake rating items. Update the rating of each fake rating item in the set of fake rating items based on the rating of the real rating items.

[0148] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. An intelligent project consulting system based on artificial intelligence, characterized in that, It includes an information collection unit (1), a consultation matching unit (2), a project scoring unit (3), a feature extraction unit (4), a project classification unit (5), a project recommendation unit (6), and a scoring update unit (7), specifically: The information collection unit (1) is used to obtain the project consultation information of the target consultation user based on the preset project consultation template, and transmit the project consultation information of the target consultation user to the consultation matching unit (2). The consultation matching unit (2) is used to receive the project consultation information of the target consultation user transmitted by the information collection unit (1), match the project consultation information of the target consultation user with the preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user, and transmit the set of candidate projects to the project scoring unit (3). The project scoring unit (3) is used to receive the candidate project set transmitted by the consultation matching unit (2) and obtain the historical scoring information of each candidate project in the candidate project set, wherein the historical scoring information includes historical scores and scoring users, and transmits the historical scoring information of each candidate project in the candidate project set to the feature extraction unit (4). The feature extraction unit (4) is used to receive the historical rating information of each candidate item in the candidate item set transmitted by the project rating unit (3), and to extract features from the historical rating information of each candidate item in the candidate item set to obtain the rating feature set of each candidate item in the candidate item set, and to transmit the rating feature set of each candidate item in the candidate item set to the project classification unit (5). The project classification unit (5) is used to receive the rating feature set of each candidate project in the candidate project set transmitted by the feature extraction unit (4), and classify the rating feature set of each candidate project in the candidate project set based on the pre-trained rating classification model to obtain the rating type label corresponding to each candidate project in the candidate project set. The rating type label includes a real rating label and a fake rating label. Based on the rating type label, the candidate project set is divided into a real rating project set and a fake rating project set. The real rating project set is transmitted to the project recommendation unit (6), and the real rating project set and the fake rating project set are transmitted to the rating update unit (7). The project recommendation unit (6) is used to receive the real rating project set transmitted by the project classification unit (5), and to recommend projects to the target consulting user based on the real rating project set, so as to obtain the project recommendation list corresponding to the target consulting user. The rating update unit (7) is used to receive the set of real rating items and the set of fake rating items transmitted by the item classification unit (5), and match the set of fake rating items and the set of real rating items based on the similarity model to obtain the real rating items that match each fake rating item in the set of fake rating items, and update the rating of each fake rating item in the set of fake rating items based on the rating of the real rating items.

2. The project intelligent consulting system based on artificial intelligence according to claim 1, characterized in that, Feature extraction is performed on the historical rating information of each candidate item in the candidate item set to obtain a rating feature set for each candidate item in the candidate item set, including: Calculate the project popularity of each candidate project in the candidate project set based on the historical rating information; Calculate the project novelty of each candidate project in the candidate project set based on the historical scoring information; The candidate project set is sorted according to the project popularity, and the candidate project set is divided into a popularity set and a non-popular set based on cross-validation. The candidate project set is sorted according to the novelty of the project, and the candidate project set is divided into a novel set and a non-novel set based on cross-validation; Calculate the popularity distribution of each popularity item in the popularity set based on the historical rating information; The non-hotness distribution of each non-hotness item in the non-hotness set is calculated based on the historical rating information. Calculate the novelty distribution of each novel item in the novelty set based on the historical scoring information; Calculate the non-novel distribution of each non-novel item in the non-novel set based on the historical scoring information; The popularity distribution, the non-popularity distribution, the novelty distribution, and the non-popularity distribution are combined to obtain the scoring feature set for each candidate item in the candidate item set.

3. The project intelligent consulting system based on artificial intelligence according to claim 2, characterized in that, The formula for calculating the project's popularity is as follows: ； in, This indicates the ratings of all users for the candidate items. Number of ratings Indicates rating user For candidate projects The rating value, when ,but Otherwise, , Represents the set of all users who rate the service; The formula for calculating the novelty of the project is as follows: ； in, This represents all user rating candidate items. Dissimilarity to other different rating candidate items express The total number of projects in China Indicates rating user The number of candidate items for scoring Indicates candidate projects With candidate projects The similarity between them; The formula for calculating the heat distribution is as follows: ； in, Indicates candidate items in the heat set The proportion of popularity to the total popularity of the entire popularity set, and , Represents a set of heat values; The formula for calculating the non-thermal distribution is as follows: ； in, Indicates candidate items in the non-hot set The proportion of popularity of non-popularity aggregates to the total popularity of [the data / intensity]. This represents a set of non-heat sets.

4. The project intelligent consulting system based on artificial intelligence according to claim 3, characterized in that, The formula for calculating the novel distribution is as follows: ； in, Represents candidate items in the novel set The proportion of the novelty of the novelty to the total novelty of the novelty set, and , Represents a novel set; The formula for calculating the non-novel distribution is as follows: ； in, Indicates candidate items in a non-novel set The proportion of novelty to the total novelty of the non-novel set. This represents a non-novel set.

5. The project intelligent consulting system based on artificial intelligence according to claim 4, characterized in that, The rating classification model employs an improved support vector machine (SVM) model. The penalty parameters and kernel function parameters of the SVM model are optimized using the Grey Wolf optimization algorithm to obtain the rating classification model. The optimization of the penalty parameters and kernel function parameters of the SVM model using the Grey Wolf optimization algorithm to obtain the rating classification model includes: Randomly initialize individuals in the gray wolf population, where the position of each individual represents the parameter combination of the penalty parameter and kernel function parameter of the support vector machine model; The positions of the individual gray wolves in the population are updated to obtain new positions for the individual gray wolves. The formula for updating the new positions is as follows: ； in, Indicates the Position at the next iteration Indicates the location of the prey. Denotes the first random parameter, and , Represents the nonlinear convergence factor. express The first random number within the interval, Indicates the distance between the prey and the wolf pack, and express, Indicates the Position at the next iteration Denotes the second random parameter, and , express The second random number within the interval; Calculate the new fitness of the gray wolf population individual at the new location and compare it with the optimal fitness of the previous iteration. If the new fitness is greater than the optimal fitness, replace the new fitness with the optimal fitness and retain the new location of the gray wolf population individual. If the number of iterations exceeds the preset iteration threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters; otherwise, the above operation is repeated.

6. The project intelligent consulting system based on artificial intelligence according to claim 5, characterized in that, Based on the set of real-rated items, item recommendations are made to the target consulting user to obtain an item recommendation list corresponding to the target consulting user, including: Obtain the set of rating users corresponding to the set of real rating items; Calculate the user similarity between the target consulting user and the set of rating users; Based on the user similarity, obtain the K-nearest neighbor set of the target consulting user; Calculate the true rating item deviation based on the true ratings corresponding to the K nearest neighbor set of the target consulting users; Calculate the predicted rating of the target consulting user for the actual rating item based on the deviation of the actual rating item; The set of real-rated items is sorted according to the predicted ratings, and the top preset number of real-rated items in the set of real-rated items are selected as recommended items to obtain the recommended list of items corresponding to the target consulting user.

7. The project intelligent consulting system based on artificial intelligence according to claim 6, characterized in that, The formula for calculating user similarity is as follows: ； in, Indicates rating user and target consulting users In classification Similarity on Indicates rating user and target consulting users Those who were also evaluated belong to the category. Project collection, Indicates user Belonging to The average rating of the items in the list, Indicates target consulting users Belonging to The average score of the items in the list; The formula for calculating the deviation of the actual rating items is as follows: ； in, Indicates belonging to the same category Project and projects The average deviation, Indicates that they belong to the same category Project and projects The set of users who rate simultaneously Indicates belonging to the same category Below and target consulting users The set of the K most similar nearest neighbors; The formula for calculating the predicted score is as follows: ； in, Indicates target consulting users For the project Predicted score Indicates excluding projects In addition, target consulting users Rated categories A collection of projects.

8. A project intelligent consulting method based on artificial intelligence, applicable to the project intelligent consulting system based on artificial intelligence as described in any one of claims 1-7, characterized in that, include: Based on a preset project consultation template, the project consultation information of the target consultation user is obtained, and the project consultation information of the target consultation user is matched with a preset project database to obtain a set of candidate projects that match the project consultation information of the target consultation user. Obtain historical rating information for each candidate item in the candidate item set, wherein the historical rating information includes historical ratings and rating users; Feature extraction is performed on the historical rating information of each candidate item in the candidate item set to obtain the rating feature set of each candidate item in the candidate item set; The rating feature set of each candidate item in the candidate item set is classified based on a pre-trained rating classification model to obtain the rating type label corresponding to each candidate item in the candidate item set. The rating type label includes a real rating label and a fake rating label. Based on the rating type label, the candidate item set is divided into a real rating item set and a fake rating item set. Based on the set of real-rated items, project recommendations are made to the target consulting user to obtain a project recommendation list corresponding to the target consulting user; The set of fake rating items and the set of real rating items are matched based on a similarity model to obtain real rating items that match each fake rating item in the set of fake rating items. The rating of each fake rating item in the set of fake rating items is updated based on the rating of the real rating items.

Images