Distributed consistent search result random sorting method

Abstract

The invention discloses a distributed consistent search result random sorting method, which comprises the following steps of: preparing search data, and generating an index primary key identifier; initializing global parameters for controlling random sorting processing; a search thread is started, and a random time factor is initialized; calculating an initial score and a random score for the matched product data; calculating the sum of the initial score and the random score as the total score of the product data; the method is not influenced by a distributed environment and a network environment, the consistency of search results in the distributed environment is supported, and the effects of guaranteeing the fairness of product search and improving the user experience of product retrieval of a user are achieved.

Description

technical field [0001] The invention relates to the field of Internet e-commerce information retrieval and processing, in particular to a distributed consistent search result random sorting method. Background technique [0002] With the development of computer technology and Internet technology, the Internet and the e-commerce developed on it have penetrated into people's daily life. In the B2B e-commerce platform, suppliers release a large amount of product information every day. The engine is retrieved. As the amount of product data on the platform increases to a certain order of magnitude, many products will be under the condition of equal matching and importance of search terms. At this time, how to ensure the fairness of the order of product search and display, so that search users can see as many as possible Relevant product information, giving users a better experience, has become an important consideration in search ranking. [0003] When searching and sorting, usu...

AI Technical Summary

Problems solved by technology

[0004] (1) Sorting according to the release time of the product publisher or the update time of product information, however, the ranking of products with an earlier release or update time is always lower, or the publisher repeatedly updates product information or updates in order to rank higher. Republishing products, causing unnecessary information updates

When all product publishers adopt the freshness strategy, on the one hand, it will cause unnecessary system update pressure, and on the other hand, the update time becomes that the publisher deliberately updates product information to improve the ranking. This design of ranking by time becomes Pointless

[0005] (2) Place the random program in the search server where the data center is located; as the scale of users on the platform gradually expands, in order to ensure service performance and high availability, the search server is deployed in multiple data centers, which is convenient for users to access efficiently nearby. There will be a phenomenon: when the user accesses different data centers, such as in Nanjing and Guangzhou, or accesses under the network provided by different network operators, such as China Telecom and China Unicom, the sorting results will be different. Users in different network environments or regions, or even the same user will find that the same search term will search in different order of results when the network changes, which results in a bad user experience

When buyers and sellers communicate with users or when users and platform operators exchange ranking results, they cannot communicate based on consistent ranking information

[0006] (3) Set random rules to move regularly within a range. However, users can perceive the random rules. For information publishers, they hope that the products they release will be in the forefront, and they will artificially use the rules. For example, when the products they release are about to roll Insert new products at the rear, do manual interference such as loading and unloading operations, and destroy random fairness

Under this strategy, when the number of products with the same matching degree and importance is greater than N, the product will not be able to participate in random ranking, and often when this happens, tail products with the same matching degree and importance cannot appear In the top N, or the number of appearances in the top N is uneven, which affects fairness

[0008] The patent "Search Engine Result Sorting Method" (patent application number: 201210435060.7) divides webpage resources with the same matching score into multiple sets according to the height of the weighted score through the retrieval program; sorts webpage resources with high matching scores in the Before the web resources with low matching scores, among the web resources with the same matching score, the set with high weighted score is sorted before the set with low weighted score, but each webpage resource is randomly sorted in the first set, and This is shown to the user; the patent uses a search program to divide the search results in a set into multiple sections, and moves the first section to the end according to a certain time interval, and advances the second and subsequent sections sequentially, so that The search results have the opportunity to be displayed fairly, but there are the following deficiencies: (1) The positions between the search results in the random sorting collection segment are not randomly sorted, and the results in the segment are not fair; (2) The segment is moved from the beginning to the end , The random way forward in the subsequent segment is regular, and may be used by users by changing data, etc., thereby gaining a ranking advantage and undermining fairness; (3) There is no guarantee that the random order between multiple data centers is consistent, and buyers and sellers access different data Communication after the center, user network drift, etc., there will be inconsistencies in the order

Images