Dairy powder HS code references

List of the contents of this article:

• 1,Dairy powder HS code references Recommended System Paper Reading (21)-DeepFM Upgraded Version xDeep FM
• 2. Overview of Recommended System Products and Algorithms丨Product Miscellaneous Series
< Li style='margin-bottom: 3px;list-style: none'>3, Recommended System (I): Item-based Collaborative Filtering Algorithm
• 4, recommended system (III): recommended algorithm based on matrix decomposition
• 5, 07_Recommended System Algorithm Detailed Explanation

Recommended System Paper Reading (21) - DeepFM Upgraded Version xDeepFM

1. Look at Figure c first. xdeepfm is sent to DNN for ctr estimate through a CIN vector concat. This paper's The key is the whole CIN. The full name of CIN is Compressed Interaction Network. Let's introduce in detail how CIN does it.

2. wide&deep First of all, let's introduce the wide&deep model. The model structure is as follows. In the model, the wide part is responsible for memory, and the deep part is responsible for extension (generalization).

3. This is Ali's mother.Mom published another masterpiece on 2020SIGIR. Let's read this paper.

4. The original collaborative filtering method ignores this kind of information, so it is not enough to embedding well when performing user and item representation.

Overview of Recommended System Products and Algorithms丨Product Miscellaneous Series

1. The recommended algorithms of many products depend on three types of data: descriptive information related to objects (such as recommended shoes, including shoes The layout, applicable object, material and other information, user portrait data (referring to user-related data, such as gender, age, income, etc.), user behavior data (such as users' browsing, collection, purchase, etc. on Taobao).

2. First, review the recommendation principle of UserCF algorithm and ItemCF algorithm: UserCF recommends items that users like with the same interests and hobbies, while ItemCF recommends users those that have similar behaviors to the items he liked before. Items.

3. A complete recommendation system usually includes 3 components: user modeling module; recommendation object modeling module; recommendation algorithm module. The recommendation system is an information filtering system used to predict users' ratings or preferences for items. It can find out the connections that will eventually occur between the user and the item.

4. The recommendation system uses e-commerce websites to provide customers with commodity information and suggestions. The recommendation system can help users decide what products to buy and simulate salespeople to help customers complete the purchase process.Personalized recommendation is to recommend the information and goods that the user is interested in to the user according to the user's interest characteristics and purchasing behavior.

5. The content of information flow is not purely recommended by algorithms, and manual operation is also an important part of it.

recommendation system (I): item-basedThe collaborative filtering algorithm

1. The collaborative filtering algorithm is the most classic and commonly used recommended algorithm. The basic idea is to collect user preferences, find similar users or items, and then calculate and recommend them. The core idea of the item-based collaborative filtering algorithm is to recommend users those items that are similar to their previous favorite items.

2. The item-based collaborative filtering algorithm is the most widely used recommended algorithm in e-commerce at present. In non-social networking websites, the intrinsic connection of the content is an important recommendation principle, which is more effective than the recommendation principle based on similar users.

3. User-based collaborative filtering algorithm: Based on the assumption that "you are also likely to like what people like with similar preferences."Therefore, the main task of user-based collaborative filtering is to find out the user's nearest neighbor, so as to make a score prediction of unknown items according to the preferences of the nearest neighbor.

4. This is a typical example of collaborative filtering of items. Collaborative filtering based on items refers to the recommendation of items based on the behavioral similarity of items (such as beer and diapers being purchased at the same time). The algorithm believes that item A and item B are very similar because most users who like item A also like item B.

5. In a personalized recommendation system, when a user A needs personalized recommendation, you can first find other users with similar interests to him, and then recommend the items that users like but user A has never heard of to A.

6. In general, association rules are classified as dynamic recommendations, while collaborative filtering is more regarded as static recommendations.The so-called dynamic recommendation means that the basis of the recommendation is and only the current (recent) purchase or click.

Recommended system (III): Recommended algorithm based on matrix decomposition

Matrix decomposition funkSVD: This matrix decomposition is not like that of the linear generation, and it belongs to pseudo-decomposition. The main idea is to replace the matrix of m*n with two m*k and k*n matrices. Because in the recommended system, the matrix is very sparse, the decomposed matrix is generally dense, and the empty value can be obtained by multiplying the row.

When the idea of matrix decomposition appears in the recommended model, SVD (singual value decomposition) naturally comes to mind. SVD can decompose a matrixIn the form of , in the main diagonal line of D, the singular values are sorted from to to small. We select the first few singular values and the vectors corresponding to U and V to achieve dimension reduction.

On the contrary, if the similarity is normalized, the coverage of the recommended system can be improved.

07_Detailed explanation of the recommended system algorithm

User-based (User-CF): The basic principle of user-based collaborative filtering recommendation is that according to all users' preferences for items, find and the current user's taste and Prefer similar "neighbor" user groups and recommend items preferred by nearby neighbors.

First of all, review the recommended principle of UserCF algorithm and ItemCF algorithm: UserCF givesUsers recommend items that users who share the same interests and hobbies like, while ItemCF recommends items that have similar behavior to the items he liked before.

Finally, a good recommendation system design can enable the recommendation system itself to collect high-quality user feedback, constantly improve the quality of recommendations, increase the interaction between users and the website, and improve the revenue of the website. Therefore, when evaluating a recommendation algorithm, it is necessary to consider the interests of the three parties at the same time. A good recommendation system is a system that can make the three parties win-win.

The coverage rate reflects the ability of the recommendation algorithm to discover the long tail. The higher the coverage rate, the more the recommendation algorithm can recommend the items in the long tail to the user. The numerator part represents the number of all items recommended to users in the experiment (set deduplication), and the denominator represents the number of all items in the data set.

Recommendation systems are usually divided into three categories: content-based recommendation algorithms, collaborative filtering recommendation algorithms and hybrid model recommendation algorithms. The content-based recommendation algorithm essentially analyzes the content of items or users to establish attribute characteristics. The system recommends information similar to the attribute features they are interested in to users according to their attribute characteristics.

This algorithm is based on the assumption that things are clustered and people are divided into groups, and users who like the same item are more likely to have the same interests. The collaborative filtering recommendation system is generally applied to systems with user ratings, and users' preferences for items are portrayed through scores.