Brief introduction:The search for most products is the continuous optimization and iteration of technical students, and it is easy to ignore the role and value of operational students who can directly access the business side. So today, I will share with you the operation of the students in the search on what play? Alibaba Cloud Open Search is a one-stop visual search development platform. What optimization actions can product/operation students participate in on the console?

Reading time: 5 minutes

Suitable group: people with search operation positions, product positions, and interested in search technology;

Search is a necessary function of every product and the most important one for business growth. The more information a product has, the more attention it will pay to search ability. In search scenario, users “take the initiative” to generate content, which requires search technology to accurately analyze recall and hit the search intention, so as to achieve the purpose of business transformation.

The search of most products on the market is the continuous optimization and iteration of technical students, and it is easy to ignore the role and value of the operation students who can directly reach the business side. So today, I will share with you the operation of the students in the search on what play? Alibaba Cloud Open Search is a one-stop visual search development platform. What optimization actions can product/operation students participate in on the console?

What operations can be done on search that?

  1. Cooperate with product technology, be responsible for search indicators, and follow up search capability iteration continuously;
  2. Output search and evaluation reports by analyzing business indicators, develop optimized solutions, and improve product capabilities and user experience;
  3. Make breakthrough exploration of user experience from product and operation level, combine search and guide function to cooperate with operation plan, improve business transformation;
  4. According to the laws and regulations, control the content security, and timely deal with the bad information content in the platform;

Operation Students Difficulties

  1. Lack of search related technical knowledge reserve, technical concept, logic is not clear, not conducive to the development of operation work and the promotion of the project;
  • How do rewrite, word segmentation, recall, and sorting work in search, and what optimizations can operators participate in?
  • What are the ways to play personalized search?
  • How to optimize search to improve user business monetization?
  1. Long demand solution cycle, difficult to respond in time and quickly, slow business development;
  2. The enterprise does not have the perfect data management ability, the operation can not search the business index data in real time to make the corresponding operation analysis and operation decision;
  • Core search operation data: search, traffic, behavior, transaction, user analysis, Query analysis, etc.
  • Personalized search guide: drop-down prompts, hot words, shading and other data analysis;

Search for core logic interpretation

Search business process:

1. Interpret user input

Function description of query semantic understanding: “translate” the query input by users into the meaning understood by the system, realize effective communication between human and computer, analyze the search intention of users, recall the most relevant content/commodities, and solve the search demands of users.

Each function in query semantic understanding will directly affect the analysis and recall effect of the user’s search intention, and thus directly affect the click-through rate, bounce rate, conversion rate and other business indicators. At the same time, the operation, product and technology should be combined with their own product situation to continuously optimize and explore.

Example: Search for “AJ1 North Carolina New Sneakers” and the computer does the following query

2. Filter content that is relevant to user intent

After reading the user’s query, you get a set of standardized words that correspond to the relevant content, and the content is filtered based on two concepts: recall rate and accuracy.

  • Accuracy refers to the percentage of relevant content found in a search;
  • The recall rate is the percentage of the content that is actually searched.

What is a recall? This process is called recall, in which word segmentation is carried out by keyword query by the user, and word segmentation is quickly located to the document by searching the inverted linked list.

When the ratio of these two indicators is closer to 1, the effect is better. However, in some cases, accuracy and recall rate are a set of contradictory indicators. For example, if only one search result is searched and the user’s true intention is found, the accuracy rate can reach 100%, but the recall rate is low. These two concepts are key metrics in search optimization and refer to higher level search mechanisms. Note: Not all results that contain the user query keyword should be recalled.

3. Sort search results

After the query analysis is recalled, the content/products that best meet the user’s intention will be reasonably sorted to increase the click rate and prevent users from jumping out. Next, look at the collation of searches.

  • Roughing: Conduct the first round of audition selection for search results. Roughing should be as simple as possible (select the most important contents of the document, such as the textual nature and timeliness of the news category) because all the documents need to be traversed. Score the documents according to the expression and sort the results.
  • Fine sorting: select TOPN items from the rough sorting results of the first round to calculate the score value of more details in the second round according to the fine sorting. The final sorting is carried out according to the score value and returned to the user.
  • Sorting expression: A mathematical expression used to control the sorting of search results documents, supporting basic operations (arithmetic, relational, logical, bit, conditional), mathematical functions, sorting characteristics, etc. Sort expression can be used for depth tuning of sorting effect.

4. Personalized search peripheral functions :(operational focus!!)

4.1 Hot search shading

Hot search shading is a basic function of a complete search engine, usually occupies an important position in the search box entry, providing indispensable business value. It is located in the upstream of the whole workflow of search engine, and plays a role in laying the groundwork for search optimization. It can greatly reduce the difficulty of tuning links such as query understanding, sorting and operation intervention, and it can have a large space to play when combined with operational strategies.

From the user’s point of view, hot search shading can generally meet the following needs:

  1. I want to wander casually, do not know what to search good, can you recommend some high-quality query words to me?
  2. I want to know what everyone is searching for. You can’t go wrong with the crowd
  3. It is best to recommend Query based on my interests, but also with variety. I want to see the content that I am interested in, and also want to explore something outside of my interests

From an operator’s perspective, hot search and shading can provide the following value:

  1. I want to know which Queries are searched the most. Popular Queries are the wind vane of users’ interest. By analyzing popular Queries, we can grasp the trend of users’ interest and provide decision-making basis for making operational strategies
  2. I want to recommend some good Query to the user. If the user has input, the drop-down prompts guide the user’s intention, but how do I recommend good Query when there is no input?
  3. If you recommend popular Queries to users, you can’t always give the hottest ones. Diversity should be taken into consideration. On the one hand, user experience should be taken into account, and on the other hand, exposure opportunities for some of the second popular Queries should be given
  4. By analyzing users’ behaviors and combining with their interests, Query is recommended, which not only gives consideration to user experience, but also can improve the business objectives with a specific purpose

Business indicators that ✅ operations can focus on:

Hot search:

  • PV: the number of requests for hot search (and returned successfully) on that day;
  • Hot-search UV: the number of users requesting overheat search that day;
  • Hot-search UV-CTR: Users’ clicks on Hot-search results;
  • PV-CTR: The number of hits of popular searches
  • Guided search PV-CTR: measures the recall and ranking effect of popular search guided search;
  • Guided Search (GMV) : measures the effect of leading searches on purchases;
  • Guided Search Favorites/Comments/thumb up Conversion Rate: measures the effect of leading searches on Favorites/Comments/thumb up;


  • Shading PV: The number of times that Shading was requested (and returned successfully) on the day;
  • Shading UV: the number of users who requested shading on the same day;
  • Shading UV-CTR: The user’s click on the shading result;
  • Shading PV-CTR: Shading click condition;
  • Guided search PV-CTR: measure the recall and sorting effect of shading guided search;
  • Guided Search GMV: measures the effect of shading on guided purchases;
  • Booted Search Favorites/Comments/thumb up Conversion Rate: Measure Shading Booted Favorites/Comments/thumb up Effect;

4.2 Pull-down prompts

The drop-down prompt is the basic function of the search service. In the process of the user entering the query word, the candidate query can be recommended intelligently to improve the user’s input efficiency and help the user find the desired content as soon as possible. You can use Chinese prefix, Pinyin full spell, Pinyin first letter simple spell query, Chinese characters plus Pinyin, prefix after word segmentation, Chinese homophones and so on to query the candidate query in the drop-down prompt.

✅ Operations can focus on drop-down tips for business metrics:

  • PV: The number of times that the pull-down prompt was requested (and returned successfully) on that day;
  • Pull-down prompt UV: the number of users who requested the pull-down prompt on that day;
  • Pull-down prompt pV-CTR: the click of the pull-down prompt to measure the effect of recall and sorting of the pull-down prompt;
  • Dropdown prompt UV-CTR: the user clicks on the result of the dropdown prompt to measure the effect of recall and ordering of the dropdown prompt;
  • Guided search GMV: Pull down the transaction amount of guided search;
  • Guided Search Favorites/Comments/thumb up Conversion Rate: Measure the effect of a drop-down prompt guiding Favorites/Comments/thumb up;

Introduction to the open search platform

OpenSearch Ali Cloud is a one-stop intelligent search business development platform built by independently developed large-scale distributed search engine. Without development, high-quality search services can be obtained by one-click access. It has built-in core search engine which has been accumulated by Ali technology for many years, and has cutting-edge search and algorithm capabilities in the industry. And fully open to support the internal call of customers’ own algorithm model, to meet the business needs of various industries and scenarios, and achieve mutual success and common growth with customers;

Operation highlights and advantages:

  1. Visual console, clear module, friendly to novice personnel, easy to operate, without waiting for the development cycle, convenient for the operation of products other than technology personnel according to the business situation at any time effect tuning;
  2. Industry leading technology to create a unique industry search template, one-button configuration, built-in industry search capabilities, without development training can have the industry attributes of high-quality search capabilities;
  3. Support the algorithm model of the developer to return to the online immediately, according to their own business situation for model development superimposed on the existing platform capabilities;
  4. A/B Test is supported to facilitate the business to allocate A certain proportion of traffic prior to full use and avoid the negative impact on online business caused by blind use.
  5. Support cloud monitoring alarm, through the cloud monitoring of the application storage capacity, computing resources, QPS query and other indicators to monitor. Help to monitor the use of the application, and support to set alarm rules for monitoring items, keep track of business trends.
  6. To support customized search service, there is no need to set up a technical team to solve business difficulties. Students from top Ali technology and algorithm engineering team will help you solve problems.

Operation/product operable menus and functions

Application Management Related

1. View: application list, application details, application specifications, application cloud monitoring;

2. Operation: Second-level expansion capacity, variable specifications, easy to cope with big promotion and other activities;

Search algorithm center

1. View: All menus can be viewed

2. Recall configuration:

A. participle management:

I. Test the effect of word segmentation

Ii. Add, delete and modify the participle entry of the custom participle;

B. Query Analysis:

I. Search tests

Ii. Configure query analysis rules: rewrite policies, function selection

Configure rewrite policy: Control the recall of terms to be included in the query results in an AND OR OR relationship. Rewriting directly affects recall results and can be adjusted according to business conditions.

Example: if Query is “Nike sneakers”, the term after participle is “Nike/sports/shoes”.

    • AND: Query (default:’ Nike ‘AND default:’ sports’ AND default:’ shoes ‘)
    • (default:’ Nike ‘OR default:’ sports’ OR default:’ shoes ‘)

Query analysis function selection: select industry template default to all selected state.

C. Dictionary management:

Intervention entries were added, deleted and modified for each functional dictionary: spelling correction, stop words, synonyms, entity recognition, word weight and category prediction intervention

3. Sort configuration:

A. Sorting policy management: configure the sorting expression according to business requirements and optimize the sorting effect

B. Search test: you can view the score results of each function

4. Search and guide function:

A. Dropdown prompts:

According to the characteristics of different industry data, the drop-down prompt function provides the corresponding optimization template. Generic, e-commerce and content industry templates are currently supported;

I. View the drop-down prompt related reports and effect preview

Ii. Select to configure the black and white list

III. Personalized configuration: high-frequency search terms, the user’s search frequency is taken as the candidate ranking basis, and the user’s search frequency which meets the recommended conditions is given priority as the drop-down query candidate. Historical search terms that give preference to queries previously searched by the user. Sort candidate query intelligently according to user behavior information such as click and purchase.

B. Hot search and shading:

I. View reports related to hot search shading and preview the effect

Ii. Select to configure the black and white list

Iii. Effect optimization – behavior data: based on the statistical search log out/background results can satisfy the demand of cold start phase, began to play a role in the function, after Suggestions associated hot search/background user click event, within the system through the act of acquisition data to further optimize effect, the behavior data has the following benefits:

  • Various indicators, such as PV, UV and fruitless rate of guided search, can be obtained by statistics to measure the use effect of this function and provide a basis for subsequent improvement.
  • It can analyze the interest trend of the user group and provide a basis for the development of operational strategies.
  • Intelligent means can be adopted to recommend Query. Data can be annotated through users’ click behavior, and models may be trained according to different optimization objectives (the default is click rate optimization). Query can be recommended through models, which has strong generalization ability.
  • Personalized recommendation of hot search terms can be made. After knowing which Queries users have clicked, targeted recommendations can be made based on users’ preferences.

Statistical reports

All reports can be viewed: business operation report, drop-down prompt report, hot search shading report, A/B test report;

All of the above menu and function introduction, operating procedures, considerations can be found in the open search product documentation.

If there are any product guide requirements, can fill in the questionnaire for expert guidance \ > >\_J8cRj

Copyright Notice:The content of this article is contributed by Aliyun real-name registered users, and the copyright belongs to the original author. Aliyun developer community does not own the copyright and does not bear the corresponding legal liability. For specific rules, please refer to User Service Agreement of Alibaba Cloud Developer Community and Guidance on Intellectual Property Protection of Alibaba Cloud Developer Community. If you find any suspected plagiarism in the community, fill in the infringement complaint form to report, once verified, the community will immediately delete the suspected infringing content.