This article describes how ES users of Alicloud can access the open source compatible version of Elasticsearch to enrich the industry thesaurus, improve their query semantic understanding ability, and obtain the same search results without development or algorithm investment.

Open Search open Source compatible edition introduction

Many customers are more familiar with open source technology stack when building search business, and choose open source engines such as Elasticsearch/Solr to take charge of search recall. However, training NLP and sorting algorithms outside recall engines is time-consuming and labor-intensive, and most of them are in a state of high investment and low return.

Elasticsearch open source compatible version of Elasticsearch engine is based on alibaba’s accumulation in the search field. The Elasticsearch engine instance of open search runs in the Elasticsearch cluster of Alibaba cloud as a plug-in. It provides alibaba’s self-developed industry word segmentation ability and Query analysis and understanding ability, thus reducing users’ input on algorithm modules and allowing them to devote more energy to business transformation and product functions.

Business flow chart:

Open search compatible edition advantage

  • Enrich Elasticsearch open source engine thesaurus based on Alibaba’s accumulation of thesaurus for many years and Dharma Academy NLP technology to improve the search effect;

On the basis of general word segmentation, industry word segmentation ability is also added, which can cover e-commerce, IT content, education, games, mutual entertainment and other industries.

  • Query understanding of Elasticsearch engine to accurately locate user search intent;

Through a series of intelligent semantic analysis on Query (spelling correction, synonym rewriting, word weight, stop words, entity recognition) to understand the user’s search intention, rewrite the user’s search Query, make the recall results more in line with the demand;

  • Provides the ability to customize the segmentation, query analysis intervention dictionary, even after the open search console configuration

Users can adjust and optimize according to their own business, effectively respond to the search demand, improve the search effect and user experience;

Open source compatible version of ES engine instance creation and configuration

Create Elasticsearch engine instance

1. Log in to the Open Search console, go to the Instance Management-ElasticSearch engine page, and click Create App:

2. Click the Elasticsearch Enhanced Sale page, select the negotiation type and region required to create the app, enter the app name, select the resource group, and click Buy Now.

3. You can view the newly created application instance on the Instance Management page of Elasticsearch Engine.

2. Configure the Elasticsearch instance

Configuring an application includes associating Elasticsearch instance with Aliccloud, installing the plug-in, and completing the configuration.

  1. Example Management -Elasticsearch engine page, find the instance to be configured, click Configure, enter the page of Associating Elasticsearch application with Ali Cloud:

  1. Install plug-in:

  1. Click “OK” to start installing the custom plug-in (which triggers restart of the Elasticsearch cluster) :

  1. If the configuration is successful, wait for plug-in installation:

Test Elasticsearch instance search

  1. Elasticsearch Custom add-on is installed on the Elasticsearch console instance details page.

  1. Log in to the Elasticsearch visual console and test the installed plug-in using the Dev Tools tool:

For more configuration details, see the product documentation:Help.aliyun.com/document\_d…

Customer case

A new retail customer, to create a 1km community online shop service, to provide users with eating, drinking, play, music integrated life services.

Customers search for business pain points

  • Poor self-built search effect, inaccurate search, search can directly affect the user experience;
  • Lack of industry thesaurus, self-research difficult, long development cycle, difficult to respond to business needs;
  • Mature search engines involve offline module, online module, query and understanding service, algorithm platform and other system components, requiring a lot of development, algorithm tuning and continuous complex operation and maintenance work, high self-construction cost;

Open source compatible version solution

1. Call the open search e-commerce industry thesaurus

The training corpus comes from millions of annotated e-commerce industry data accumulated by Taobao search for years, which can accurately identify e-commerce attributes such as commodity brand, category and product characteristics.

2. Invoke the e-commerce query semantic understanding function

  • E-commerce spelling correction

The query entered by the user is not always correct. Incorrect input may cause the query result to be incorrect or null. Therefore, spell-check is required for the user’s input. The spelling check function provided in OpenSearch query analysis can correct the errors in the query words and give the correct query words. And according to the credibility of error correction, decide whether to use the corrected word for the current query.

  • Synonyms for E-commerce

The synonym function is to extend the synonyms of the query words and expand the recall of the documents with the same meaning of the query words.

  • E-commerce entity recognition

Named Entity Recognition (NER) refers to the Recognition of semantic entities with specific meaning in query words. According to the recognition results, the query words are rewritten according to the weight of the entity type to make the recalled documents conform to the query intention.

Effect of feedback

Without additional input of human resources and without changing the usage habits of the existing ES, high-quality search results can be obtained in just 15 days from learning about the test to access and putting online, and the enterprise has more resources and energy to devote to product functions and business improvement.

  1. Product search results rate dropped from 30% to less than 5%, the index is still in continuous optimization;
  2. Search-led conversions increased 7%;
  3. CTR of merchant store search increases by 5%, which will directly affect merchant check-in and advertising revenue.

The original link

This article is the original content of Aliyun and shall not be reproduced without permission.