FAQ

Updated at:

General search

  1. What are the differences between the search APIs: GenericSearch - Standard Search API, AISearchV3 - Multi-stage Streaming API, and GenericAdvancedSearch - Enhanced Search API (Coming Soon)?

    GenericSearch and AISearchV3 are equivalent. AISearchV3 uses the streaming Server-Sent Events (SSE) format. GenericAdvancedSearch is an enhanced API. It retrieves authoritative and diverse web pages to provide more search results.

  1. Can I adjust the pagination and page size in the search structure? How do I retrieve more results?

    Pagination is supported through the `page` parameter. The number of results per page cannot be adjusted. To retrieve more results, you can make multiple API calls and increment the `page` parameter in each call.

  1. Why do the results sometimes contain more or fewer than 10 items instead of exactly 10?

    Yes, this is due to two filtering mechanisms. First, some sites in the recall results are filtered out. Second, the reranking process removes sites with a relevance score below 0.3, which can result in fewer than 10 items. More than 10 items are returned when a news card (`news_uchq`) is recalled. A single card contains 10 results, and when expanded, the total can be up to 19.

  1. When I run the Python sample code, I receive the error: AttributeError: 'NoneType' object has no attribute 'get_credential'. Why?

    This error occurs because `access_key_id` or `access_key_secret` in your configuration is empty. Make sure your environment variables are set correctly.

  1. When I run the Python sample code, I receive the error: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate. Why?

    This error is caused by an expired SSL Certificate. You can delete the old certificate from your local machine and reinstall the latest version of the `certifi` package.

  1. After I place an order, why does the console always show the status as "Activating"?

    For the activation procedure, see Web Search Activation Instructions. After you place an order, a 15-day trial account is activated by default. The console shows the status of the official account, not the trial account. You can call the API for testing during this time. To convert to an official account, contact us.

  1. What are the best practices for query terms?

    The open-domain search engine is optimized for keyword searches. Long queries can reduce search precision. When you send a request, you can use a model to rewrite complex questions into one to four keywords. You can also rewrite one query into multiple queries. For best results, keep the query length under 15 characters. This improves both response time (RT) and relevance.

  1. How timely are the index updates? Can the index include content published on the same day?

    Web pages are updated at different frequencies. Most pages are updated daily, while news and information pages are updated every few minutes. Hot topics are retrieved using a vertical API (`SceneItem`). Some pages are only updated weekly or monthly.

  1. After upgrading from a trial account to an official account, do I need to change my AccessKey, API-Key, or code?

    No, you do not. Upgrading from a trial to an official account only changes your account status. If you use the same Alibaba Cloud account to place both orders, you do not need to change your code, AccessKey, or API-Key. After the upgrade, standard metering and billing will apply, and you will receive daily bills on a pay-as-you-go basis.

  1. Can I access the service over a private network from an Alibaba Cloud data center?

    Yes, you can. If your client is in an Alibaba Cloud VPC, you can use a VPC endpoint to access the service. Endpoints are supported in eight regions. For more information, see Service Endpoints.

  1. The recall results for some time-sensitive queries are poor. Examples include "What is the weather in Shanghai today?", "NVIDIA stock price", and "How much is 200 USD in CNY?".

    It is difficult to obtain results for these types of queries through web page indexing. We supplement the recall using a vertical-specific API. For more information, see Scenario-based Calling of SceneItem Overview. If a recall is available in SceneItem, you can prioritize it by directly passing these results to the model's context. Currently, scenarios such as weather, time, and date are supported. More scenarios will be added in the future.

  1. Why does the publishTime in the results not match the time on the actual linked page?

    The publication time of a web page is predicted by an algorithm. In rare cases, this prediction can be inaccurate. We are continuously working to improve its accuracy.

  1. Why does the mainText in the recall results differ from the main text on the actual linked page?

    This can occur because some web pages contain dynamic content. The `mainText` field returns a cached result that is updated periodically. In rare cases, the content may be out of sync with the live page due to differences in update timing. We plan to optimize this in early April. In the long term, many of these discrepancies will be resolved using the `SceneItem` vertical API.

Related queries