618 Technical Special (1) Unknowingly, it is 3 times the budget. Why can't you stop buying and buying?
Abstract: is behind our lack of self-control, or the e-commerce platform is too good at understanding people's hearts, we might as well take a look at it from the technical dimension.
This article is shared from the HUAWEI CLOUD community " 618 Technical Special (1) Unknowingly over budget 3 times, why can't you stop buying? ", original author: technical torchbearer.
It's 618 again in the year. When everyone can't help buying, buying and buying, will they have such a feeling: as if there is a magical force pulling themselves, once the goods are added to the shopping cart, they can no longer stop.
The same is true when watching live broadcasts. I heard the anchor yelling in his ear: buy one and get another, buy two and get three, and place the order with a shake of your hand.
Is it our lack of self-control or the e-commerce platform is too good at understanding people's minds? We might as well take a look at it from a technical perspective.
Graph database: Establish relationships between entities and gain insights into your preferences
Xiao Wang of Chrysanthemum Factory is a novice dad, ready to show off in 618 and buy some milk powder bottles for his baby. Xiao Wang opens an e-commerce app, adds milk powder to the shopping cart, and slides down. He sees that this little toy is not bad, and that set of clothes is also very cool... The operation is as fierce as a tiger, the more he buys, the more excited he gets, and finally buys The things are three times more than what he planned.
Yes, during the big promotion, Xiao Wang accidentally fell into the recommendation network of the e-commerce system. APP homepage, shopping cart page, product detail page...recommendation system is everywhere. The backstage of e-commerce APP has already prepared for him based on the user portrait (gender, age, shopping history, search history, etc.) that he has mastered. Series of goods.
In the field of e-commerce, the recommendation system is powerful. It allows users to spend more time browsing products to achieve the purpose of increasing the customer unit price. The used behind it is the graph database technology .
People who are new to the image database can easily be misled by its literal meaning, thinking that it is a database for storing images, but it is not. are two completely different concepts, the 160d2a3b9e9fb3 graph database refers to an online database management system that stores data in a graph structure.
For example, Xiao Wang and Xiao Li from Chrysanthemum Factory are colleagues and both like to play table tennis. Here, Xiao Wang and Xiao Li are each entity representing a point. The relationship between Xiao Wang and Xiao Li—colleagues—is the edge connecting two points. This is a simple graph structure.
Through the graph structure, we can model various related scenes, from the recommendation system of social networks and e-commerce platforms to the transportation system of the entire city.
For example, the e-commerce platform will tag Xiao Wang based on his characteristics (programmer, dad, love to play table tennis, etc.), and use the tags to determine user attributes: Xiao Wang usually likes to shop for brand A digital products and is used to it Buy B brand baby products. However, Xiao Li, who also often reads brand A mobile phone information, will not buy brand B products. With the help of the graph structure, he can find out this association and realize precision marketing.
If you want the graph structure to find this relationship, you have to rely on the query analysis, calculation, storage management, and visualization of graph computing. For example, the graph database Neo4j is good at real-time query of graph data; the graph engine focuses on using mature graph algorithms for offline analysis and mining in massive graph data.
HUAWEI CLOUD's Graph Engine Service (GES) provides query and analysis services for relation-based graph structure data. Taking Huawei Mall as an example, real-time product recommendation can be achieved with the help of GES. It analyzes and compares the preferences of target users and other users, and after finding similarities, recommends the products purchased by these other users to the target users.
One of the reasons why GES can achieve a large number of real-time recommendations is EYWA, which provides a complete set of solutions from the underlying graph storage and management, the core high-performance computing engine, to the upper-level graph analysis and graph query. .
Technically, EYWA has made these optimizations:
- Distributed optimization of the graph computing framework of Parallel Sliding Window (PSW) to efficiently load graph data to meet the large-scale computing needs of the business;
- Taking into account the efficiency of graph calculation and point query, develop block data organization based on edge-set to organize data reasonably;
- Use edge set prefetch strategy to hide disk IO operations and slack BSP model to hide communication IO, thereby improving performance;
Another great hero is the graph scene and graph optimization algorithm owned by GES. Take the Pixie algorithm as an example. Pixie is Huawei Cloud’s attempt to build multiple data into the same graph and configure it on this heterogeneous graph. Algorithms designed based on the corresponding schema, point and edge attributes, and weights. It is a brand-new real-time recommendation algorithm that overcomes the data acquisition and fusion problems of heterogeneous graphs, supports comprehensive recommendation under multiple request nodes, and can meet the needs of various complex, time-varying, and diverse recommendation scenarios.
Therefore, the real-time recommendation algorithm provided by the GES graph engine service performs recommendations under the combined effect of multiple relationships (historical interaction information between users and commodities, potential relationships behind people, commodities and commodities, etc.), with higher accuracy; with large amounts of data, It can still achieve a good real-time recommendation effect and has strong scalability. For specific data, please refer to this article: " HUAWEI CLOUD New Generation Black Technology Core Algorithm Secret ".
Knowledge Graph: Michelin chefs "process" data raw materials, intelligently recommend fast, accurate, and ruthless
If the graph database emphasizes the storage, query and management of "data raw materials", then the knowledge graph is the chef of the Michelin restaurant to further process the data. The knowledge graph based on the graph engine service integrates various heterogeneous and heterogeneous data to form a large-scale knowledge base to support business applications and make search results more accurate.
Based on the graph database, the knowledge graph expands the associated attributes of various commodities (various dimensions of the commodity, the preference of these dimensional attributes, the social evaluation of the commodity, etc.), due to the expansion of the hidden relationship between users and commodities, The interaction data between the user and the item is supplemented, and the person-thing association is realized based on the person-person association. Therefore, the recommendation effect can be further improved.
For example, users with the same attributes may be interested in the same type of items. When Xiao Wang, who loves playing table tennis, buys quick-drying clothes of Brand A on an e-commerce app, then when his colleague Xiao Li, who also likes table tennis, opens the app , The first one might be the shop that Xiao Wang just bought. You see, such a recommendation system based on the attributes of acquaintances has been quietly staged automatically through the knowledge graph.
The construction of knowledge graph on the one hand improves the accuracy of personalized recommendations on e-commerce platforms, and it can also be applied to intelligent customer service to help establish knowledge system cards and realize intelligent question and answer.
For example, when you enter: "display" in the customer service dialog of an e-commerce company, the knowledge card will list product introduction, features (screen size, pixels, etc.), style classification, suitable people, suitable occasions, and production processes.
The online customer service of Huawei Mall is also a typical application. It can handle all kinds of inquiries from consumers before and after sales, and reply to the product promotion information you need in a timely manner.
Moreover, based on the knowledge graph constructed by the product, it completes the question and answer more accurately than the ordinary FAQ system, and realizes the ability of product comparison, product FAQ support, and attribute query.
So, how does knowledge graph liberate manpower and make intelligent customer service so outstanding?
The mainstream knowledge graph construction method in the industry is based on the internal and public data of the enterprise. Graph service providers help customers customize the construction of knowledge graphs in the form of solutions. This method is costly and inefficient, and the production cycle is long. After an e-commerce company completes the knowledge map of a certain product, the main product may be out of date and cannot be sold.
In order to provide pipelined graph construction capabilities, Huawei Cloud Knowledge Graph Cloud Service abstracts graph construction into: ontology construction, data source configuration, information extraction, knowledge mapping, and knowledge fusion.
Because each process module is abstracted into a plug-in form, and the graph construction task is generated through combination configuration, only the plug-in configuration needs to be modified to complete the construction of enterprise knowledge graphs in different fields. At the same time, based on the pipeline design, the knowledge graph cloud service can complete the update operation under the premise of only modifying the data source, which is very suitable for the knowledge graph that needs to be updated frequently. As for how it constructs ontology, configures data sources, completes information extraction and knowledge fusion, this article is limited in length. For details, please refer to " Frontier Technology of 160d2a3b9ea55e: Knowledge Graph Construction Process and Method ".
In summary, from graph database to knowledge graph to intelligent customer service, every year at 618, the products and services recommended by e-commerce platforms for you are more and more satisfying. Intelligent customer service can also understand you in just two or three sentences. This is also the reason why consumers are caught in the shopping carnival.
618 Technical Special (2) Why is it getting easier and easier for millions of people to place orders at the same time? When consumers are captured by the e-commerce recommendation system, how do they ensure that you can buy your favorite products anytime and anywhere during the big promotion period, and how hundreds of millions of transaction data are circulated in an orderly manner to ensure that you can both Can you receive the goods in time if you grab it? This article will decipher it for you one by one.