| In recent years, the Internet is gradually overthrowing the traditional industries with business modes represented by "Online To Offline (020)". Many traditional industries start to provide users with high-quality and more personalized services by using the Internet. For example, a restaurant could largely improve users’ experience and increase its competitiveness, which supports020mode and provides customers with nutrient and personalized menus. The system built by personalized nutrient recipe recommendation in this thesis is exactly a system which can easily provide users with convenient personalized nutrient recipe. The rise of online meal ordering makes the system of great application value.The personalized nutrient recipe recommendation system in this thesis includes three parts:the front desk to show the Demo, the core algorithms and backend database. The core algorithm, namely personalized nutrient recipe recommendation method, includes data acquisition algorithm, nutrition arrange algorithm and personalized recommendation algorithm. This thesis carried out research on the design of system and core algorithm, the main work is as follows:1. Putting forward a focused web crawler based on double-queue sorting and model self-learning to crawl Internet information related to a specific topic. Based on the existing focused web crawler framework, the crawler uses double ordered queues and depth attribute of the URL to enhance the performance of search strategy module, and uses the bloom filter to optimize the module of removing duplicate URL (Uniform Resource Locator). The data analysis module is optimized by separate calculation of the topic correlation of parent-page and subpage. A new module of model self-learning is added as well. Then we gives the comparative experiments of this crawler with the general web crawler and the double queue topic web crawler, and the result shows that the harvestRate of this crawler is much higher.2. Putting forward a genetic algorithm based on L domination and team competition to solve multi-objective optimization problems in high-dimensional space. This algorithm puts the non-dominated sorting genetic algorithm II as the basic framework, and uses L dominated method and group competition based on crowding distance to get non-dominated sorting, selects the best solution for genetic operation from the parent generation to get child generation, and gets non-dominated sorting again after merging the parent generation with child generation. Then we give the contrast experiments of the algorithm with two popular multi-objective evolutionary algorithm, and the result shows that this algorithm performs significantly better than the other two multi-objective evolutionary algorithm in solving high-dimensional multi-objective optimization problems.3. Putting forward a collaborative filtering algorithm based on similarity extension and interest degree scaling to predict different personal interest of new and old users. This algorithm puts collaborative filtering based on the items as the framework, uses disgust similarity to expand item similarity and then calculates the degree of the users’ interest in the items, and uses the preference factor to scale the different interest of new and old users in items of different popularity. Then we give the adjusting parameter experiment of this algorithm and the contrast experiments with the common collaborative filtering algorithms, and the result shows this algorithm effectively promotes recommendation accuracy, recall rate,coverage and reduces the popularity.4. At last, the above three algorithms are used to constitute a personalized nutrient recipe recommendation method and design the corresponding system. The thesis introduces the overall architecture of the system, the front desk Demo and back-ground data storage structure. The specific application and the related parameters of the proposed topic web crawler, multi-objective evolutionary algorithm, and personalized recommendation algorithm are discussed, too. |
PDF Full Text Request