E-commerce recommendation algorithm

Source: Internet
Author: User

Tags: discount also has content-based recommended Apriori algorithm for easy add ASC recommended get

First, e-commerce recommendation algorithm Brief

At present, a lot of e-commerce models are B2B,B2C,O2O, in this paper and the need to illustrate the local business-to e-commerce model.

E-commerce recommendation according to the recommended content is divided into items recommended, business recommendations; Popular recommended applications mainly have three aspects: 1) for the user's browsing, search and other behavior of the relevant recommendations; 2) according to the shopping cart or item collection of similar items recommended; 3) according to historical members purchase behavior records, Use the referral mechanism for email push or affiliate marketing. The recommended algorithms are mainly divided into the following categories:

1. User-based collaborative filtering recommendation algorithm

A. Find a user collection that is similar to the interest of the target user

B. Find items in this collection that the user likes, and the target user has not heard of, recommend to the target user

2. Collaborative filtering recommendation algorithm based on item (project)

A. Based on the user's interest in a product, find the most similar items.

B. Recommend the most similar items to the target user.

Examples of collaborative filtering: Four user ABCD, the interest in 5 items ABCDE the following table (actual user interest degree of the item is different, need specific scoring quantification), here convenient to understand the principle, with two yuan value indicates whether the user is interested in the item.

0

?

 

A

P>b

C

D

E

Target items

A

1

1

0

1

0

1

B

1

0

1

0

0

C

0

1

0

0

1

1

d

0

1

1

1

0

User-based collaborative filtering: refers to the user's interest in each item to calculate the similarity, similarity algorithm has a lot of algorithms (mainly strings similarity, correlation similarity and European distance, etc.), the above and user D similar to the highest user A, user A to target items of interest is 1, you can target items to user D.

Collaborative filtering based on item: refers to the item dimension, according to the user's interest in each item, calculate the similarity between items, can calculate the item B and the object of the similarity of the most, user D is interested in item B, it is likely to be interested in the target user.

3, Content-based recommendation algorithm

The product is the objective body, extracts the characteristic of the commodity object, looks for the similarity ratio larger item to recommend. The system first models the properties of an item, and, by similarity, finds that items A and B have a higher similarity, or that they belong to similar items. The system will also find that a user likes item A, which concludes that a user may be interested in item B, so the item B is recommended to the user.

The content-based recommendation algorithm is easy to understand, mainly used in classification, clustering algorithm, the user's interest can be very good modeling, and through the increase in the property dimension of goods, to obtain better recommendation accuracy. However, the property of the object is limited, it is difficult to get more data properties, and for some items attribute feature extraction is sometimes difficult, only consider the characteristics of the item itself, ignoring the user's behavioral characteristics, there is a certain one-sidedness, for the new user has never purchased a cold start problem, can not be recommended for new users.

4. Recommendation algorithm based on association rules

The recommendation based on association rules is based on the association rules, the purchased goods as the rule head, the rule body as the recommended object. Association rules mining can find the relevance of different goods in the sales process, the association rule is in a transaction database to count the purchase of commodity set X trading in a large proportion of the transaction and purchase the commodity set Y, its intuitive meaning is the user in the purchase of certain items when the tendency to buy some other goods, Recommend items according to the association rules with higher confidence in a commodity.

According to the user's purchase record, extracting the association rule, the commonly used algorithm has the Apriori algorithm, in order to extract the frequent itemsets and the Certain Confidence Degree Association rule. The main principle of the Apriori algorithm is that if itemsets A is frequent, its subsets are frequent. If Itemsets A is infrequent, then all of the parent sets that include it are infrequent, simplifying the complexity of the selection of frequent itemsets.

5. RFM-based recommendation algorithm

6. Recommendation algorithm based on demographic characteristics

This is the simplest of the recommended algorithm, it is simply based on the basic information of the user to discover the relevance of the user, and then the similar user favorite other items recommended to the current user. The system will first model according to the user's attributes, such as the user's age, gender, interest and other information. The similarity between users is calculated based on these characteristics. For example, the system finds that users A and C are more similar by calculation. I would recommend a favorite item to C.

The advantage of the recommendation algorithm based on demographic characteristics is that there is no need for historical data, no new user cold start problem, not dependent on the property of the item, the lack of the algorithm is rough, the effect is difficult to satisfy, only suitable for simple recommendation.

7. Hybrid recommendation algorithm

Fusion of the above methods, in a weighted or series, parallel and other means of integration. The most practical applications are the recommended combination of content recommendations and collaborative filtering. The simplest way is to use content-based approach and collaborative filtering recommendation method to produce a recommended prediction results, and then use a method to combine the results, such as weighting, transformation, mixing, feature combination, cascading, feature expansion, meta-level and so on. One of the most important principles of portfolio recommendation is the ability to avoid or compensate for the weaknesses of their recommended technologies after they are combined.

1) Weighted (Weight): Weighted multiple recommended technical results.

2) Transform (Switch): Depending on the problem background and the actual situation or requirements to decide the transformation using different recommended techniques.

3) Mixing (Mixed): At the same time using a variety of recommended technology to give a variety of recommendations to provide users with reference.

4) feature combination (Feature combination): The combination of features from different recommended data sources is used by another recommendation algorithm.

5) Cascade (Cascade): first with a recommended technology to produce a rough recommendation, the second recommended technology based on this recommendation to further make more accurate recommendations.

6) Feature expansion (featureaugmentation): a technique that generates additional feature information embedded in the feature input of another recommended technology.

7) meta-level (meta-level): The input of a model that is produced using a recommendation method as another recommended method

Ii. advantages and disadvantages of various recommended algorithms

Recommended method

Advantages

Disadvantages

Collaborative filtering recommendations

Novelty interest discovery, do not need domain knowledge;

Improved performance over time;

Recommended personalization, high degree of automation;

Ability to handle complex, unstructured objects

sparse problem;

Scalability issues;

New user issues;

The quality depends on the historical data set;

The recommended quality is poor at the beginning of the system;

Content-based recommendations

The recommended results are intuitive and easy to interpret;

No domain knowledge required

New user issues;

Complex attributes are not handled well;

To have enough data to construct the classifier

Rule based recommendation

To discover new points of interest;

No domain knowledge

Rule extraction is difficult and time-consuming;

The synonym of product name;

Low level of personalization;

Based on demographic statistics

No historical data required, no cold start problem;

Does not depend on the attributes of the item, so problems in other areas can be seamlessly connected

The algorithm is rough, the effect is very difficult to be satisfied, only suitable for simple recommendation

Iii. Summary of recommended algorithms

Due to the advantages and disadvantages of various recommended algorithms and adaptation scenarios, the system begins to be different from the recommended algorithms when the system matures. At the beginning of the system, the user data is not enough, the transaction behavior records less data, if the use of content-based and collaborative filtering recommendation algorithm There are many new users cold start problem. When the system is mature, the user transaction data is more, some algorithms use matrix, produce the large sparse matrix data, the computation amount is large, need to combine the combination recommendation method. This paper summarizes the recommendations of the recommended algorithm for the business e-commerce platform at the beginning and the system maturity:

Recommended ways to use the system at a glance:

1, based on population statistics, hot search, browsing records

Based on demographic recommendations: by registering and inquiring about the attributes of some users, such as age, city of residence, education, gender, occupation, etc., can get the similarity of attributes between users;

Hot Search: Station heat search, according to the ranking to recommend;

Recommendations based on the content of the browsing record: Some product content features better extraction, such as with text description of the product, there is more difficult to extract content features, tablets, or browse the product is unknown, you need to manually or intelligently crawl related information. In general, this part of the recommendation is based on user browsing content, by extracting features, computing similarity, recommend similar products (similar products recommended accuracy may be difficult to meet the requirements, by increasing the granularity of the category recommendation is a common practice).

2. Label system

The use of tags can only be used to improve the recommended accuracy of users with a small amount of behavior, for pure cold start users, it is not helpful, because these people have not played any label. The system can also label the product, but there is no personalized factor, the effect will be a discount. In this sense, it is important to use tags to recommend, motivate users to tag, and guide users in choosing the right tags. It is also a common way to guide users to tag and classify by tags.

3, the use of multidimensional data

Everyone is in a huge social network, there are behavioral data on multiple sites, a considerable proportion of users have the habit of cross-shopping, the integration of these network data, in particular, the identity of each node to know the corresponding relationship, can bring enormous socio-economic value. Cross-domain recommendations can be achieved using the ' migration learning method '. The use of multidimensional data can solve the cold start problem of new users.

China business Bridge users from the original user base should be very large, from other data interfaces to obtain data sources, to obtain the user's basic information.

Recommended methods to use when the system is mature:

1. Collaborative filtering recommendation method

2. Content-based recommendation method

3, based on association rules, the association between user and user, the relationship between goods and goods

4. Combination recommendation method (collaborative filtering and content-based recommendation combination)

These three recommendations in the e-commerce system in the mature application of more, behavioral data sufficient to make these algorithms better recommendations, but in the data magnitude is particularly large when the data sparse problem, the general solution is to use these commodity information coarse granulation, for example, only consider a category, the data will immediately become dense. If you can calculate the similarity between categories, you can help with category-based recommendations.

Iv. Evaluation index of recommended algorithm

accuracy, diversity, novelty and coverage. Each category under the jurisdiction of many different indicators, such as the accuracy of indicators can be divided into four categories, respectively, the accuracy of the prediction score, prediction score correlation, classification accuracy, ranking accuracy four categories. The second level is the key performance indicators of commercial applications, such as the conversion rate affected by the recommendation, the purchase rate, the customer unit price, the number of purchases and so on, the third level is the user's real experience, pay attention to protect user privacy.

E-commerce recommendation algorithm

Contact Us

The content source of this page is from Internet, which doesn't represent Alibaba Cloud's opinion; products and services mentioned on that page don't have any relationship with Alibaba Cloud. If the content of the page makes you feel confusing, please write us an email, we will handle the problem within 5 days after receiving your email.

If you find any instances of plagiarism from the community, please send an email to: info-contact@alibabacloud.com and provide relevant evidence. A staff member will contact you within 5 working days.

Tags Index: