Learn to Rank

Learn to Rank is a machine learning technique that allows you to train a model to rank search results. The model is trained using a set of features and a set of relevance judgements. The model is then used to rank search results.

In this guide, you will learn how to integrate Learn to Rank with Metarank and Searchkit. You can learn more about Learn to Rank with Metarank in the Metarank documentation (opens in a new tab).

We assume in this guide that you have:

Installed Metarank and Searchkit
Trained a model using Metarank
Setup the features and relevance judgements for your model

Changes to Searchkit UI

Metarank requires session id and user id to be sent with each search request. One way to do this is in the @searchkit/instantsearch-client and provide the session id and user id via headers:

const searchClient = Client({
  url: "/api/search",
  headers: {
    "x-session-id": "my-session-id",
    "x-user-id": "my-user-id",
  },
});

Changes to the Searchkit API

To integrate Learn to Rank with Searchkit, we will use the hooks for beforeSearch and afterSearch. These hooks allow us to modify the query and the results before they are sent to Elasticsearch and returned to the UI.

We will use the afterSearch hook to send the results to the Metarank API and update the results with the new order.

export default async function handler(req: NextApiRequest, res: NextApiResponse) {
 
  const headers = req.headers
  
  const results = await client.handleRequest(req.body, {
    hooks: {
      afterSearch: async (searchRequests, searchResponses) => {
        // Get the first search request
        const [uiRequest, otherRequests] = searchRequests;
        // Get the first search response
        const [uiResponse, otherResponses] = searchResponses;
 
        // Get the session id and user id from the headers
        const sessionId = headers["x-session-id"];
        const userId = headers["x-user-id"];
 
        // Get the query from the first search request
        const query = uiRequest.body.query;
 
        // Get the results from the first search response
        const results = uiResponse.body.hits.hits;
 
 
        // Send the request to the Metarank API
        const response = await fetch("http://localhost:8080/rank/xgboost", {
          method: "POST",
          headers: {
            "Content-Type": "application/json",
          },
          body: JSON.stringify({
            session: sessionId,
            user: userId,
            fields: [ 
              { name: "query", value: query }
            ],
            timestamp: (new Date()).getTime(),
            items: results.map((item) => {
              return {
                id: item._id,
                // an optional array of extra fields that you can use in your model
                fields: [
                  { name: "title", value: item._source.title },
                  { name: "description", value: item._source.description },
                  { name: "url", value: item._source.url },
                ],
              };
            }})
          }),
        });
 
        // Get the response from the Metarank API
        const metarankResponse = await response.json();
 
        // Get the items order from the response
        const items = metarankResponse.items;
 
        // Update the order of the results for the first 20 results
        const updatedResults = items.map((item) => {
          const result = results.find((result) => result._id === item.id);
          return result
        });
 
        // Update the results in the first search response
        const updatedUiResponse = {
          ...uiResponse,
          body: {
            ...uiResponse.body,
            hits: {
              ...uiResponse.body.hits,
              hits: updatedResults,
            },
          },
        };
 
        return [
          updatedUiResponse,
          ...otherResponses
        ];
      }
    }
  });
 
}

User-restricted access to data Migrating from Algolia to Elasticsearch