# Overview of Conversational Search Assistant features The conversational search assistant analyzes user input and contextual information through a process of **proactive reasoning** and **tool\-calling**. It can search for item data, document resources, and other information within the platform, synthesize the retrieved content, and ultimately generate a response that precisely fulfills the user's request. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/36a25052dc31491ca3068cb17cccfd71 =3986x) For image & text datasets and video applications, the Conversational Search Assistent supports the following scenario features: | |**Image and text applications**|**Video applications**|\ | |> Applications associated with graphic and text datasets |> Applications associated with video datasets | |---|---|---| |**Knowledge data sources** |* **Image and text dataset:** Linked to the application’s graphic and text dataset, allowing the agent to search, read, and summarize both text and image information within the dataset.|* **Video metadata:** Metadata stored in the video dataset linked to the application, such as video titles, summaries, cast, and directors|\ | |* **Document dataset:** Linked to the document content in the application’s document dataset. Useful for supplementing industry knowledge, brand information, platform policies, and other information during the assistant’s Q&A process. |* **Video audio\-visual text data:** Derived from actual video files after processing by Large Models (LLMs). This includes **audio\-visual\-text "slices"** (segmentation). In scenarios involving multi\-level media data with series and episodes, the model understands hierarchical relationships to enable cross\-episode understanding under the same series|\ | | |* **Document data:** Document content linked to the application's document dataset Provides supplementary information such as industry knowledge, brand knowledge, and platform policies during assistant Q&A sessions | |**Supports Q&A scenarios** |**E\-commerce scenarios:** |**Video platform scenarios:** |\ | |||\ | |* **Product recommendation:** Analyzes user needs and recommends products on the platform, including general category recommendations and scenario\-based suggestions|* **Conversational Video Recommendation:** Parses complex user requests (e.g., *"Find me a sci\-fi movie with a melancholic atmosphere and a minimalist aesthetic"* ) to recommend matching video assets.|\ | |* **Product content Q&A:** Answers user questions about product parameters, functions, and usage based on the product dataset|* **Plot Interpretation & Q&A:** Analyzes video content to answer user queries regarding **plot development, character relationships**, or **key narrative events. ** For every answer provided, the assistant returns **high\-light fragment positioning**, allowing users to jump directly to the relevant scene.|\ | |* **Shopping guides:** Combines information from the product dataset with platform and industry knowledge to provide users with "how to choose" and "which is better" style guides and Q&A ||\ | | |**Video course scenarios:** |\ | | ||\ | | |* **Curriculum Search & Recommendation:** Analyzes the user's learning objectives or knowledge gaps to recommend the most suitable course series or modules.|\ | | |* **Course Content Q&A:** When users ask about specific knowledge points, the Agent generates responses based on the **multimodal content** (audio, visual, and captions) within the course videos.|\ | | |* **Precision Timestamping:** The assistant provides **exact timestamp links** to the corresponding sections within the video, enabling a "point\-to\-point" learning experience. | # Experience and configure the conversation search assistant Once there is an active dataset in the application, you can start configuring the conversation experience. Go to the **Application \> Experience Center \> Conversation Experience** tab in the console to start configuring and testing the conversation. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/fa8aa519b5674b9a9b8cb2d69a6c5e80 =2928x) # Triggering Search within Conversations During a conversation, when a user's intent requires retrieving specific item information from your datasets to provide an answer, the Conversational Search Assistent **automatically ** **initiate** ** a search request**. It intelligently queries all datasets associated with the application, retrieves relevant results, and synthesizes them into a smart, context\-aware response. **Strategy Inheritance:** When a search is triggered, the system utilizes the **Search Strategies** you have previously configured in the **[Search Experience]** . These settings directly determine the search quality, which in turn impacts the assistant's ability to satisfy user intent and ensure accurate information delivery. Please consider the following association strategy configurations when setting up conversations: * Configure reasonable data indexing in the application: see [Configure search strategy](/docs/85296/1873485)

Viking AI Search