Start your environment by clicking Start Lab above.

The environment should begin to load immediately as indicated by the Vocareum loading symbol. Please do not click Start Lab again. It may take a few mintues for the Fusion environment to fully display.

When the Fusion Login page displays, login:

  • USERNAME: admin
  • PASSWORD: password123

In this lab you will ingest and transform data with the Parallel Bulk Loader (PBL), prepare data with spark shell, and verify successful data load! Let's get started by creating a Fusion App.

  1. Click on Create new app, name it Labs and click Create App

  2. Click on your newly created app to enter the Fusion workspace, this is where you can use the PBL jobs to populate data and signals in the collections

    Ingest and Transform your Data

  3. Hover over the COLLECTIONS icon in the sidebar, then click Jobs

  4. Click Add+, from the drop-down menu, select Parallel Bulk Loader

  5. Toggle on Advanced. Let's add some parameters to ingest the data:

    • Update the Spark Job ID to BestBuy_catalog
    • Update the Format to parquet
    • Update the Path to gs://training-ecommerce/catalog
    • Change the Output Collection to Labs
    • Copy the Transform Scala below and paste it into the Transform Scala text box
    

    def transform(inputDF: Dataset[Row]) : Dataset[Row] = {
inputDF.filter("department IN ('ACCESSORIES', 'APPLIANCE', 'COMPUTERS')")
}

    <script> function copyEvent(id) { var str = document.getElementById(id); window.getSelection().selectAllChildren(str); document.execCommand("Copy") }

  6. Close the Transform Scala text box. Click Save, then click Run, then click Start

    Note: Success! The job will take up to three minutes, when it's done, the "running" icon will change to a "sunshine".

  7. Click Save

  8. Let's add another PBL job. Click Add+, from the dropdown menu, click Parallel Bulk Loader

  9. Click Advanced. Let's add some parameters to ingest the data:

    • Update the Spark Job ID to BestBuy_signals_labs
    • Update the Format to parquet
    • Update the Path to gs://training-ecommerce/signals
    • Change the Output Collection to Labs_signals
    • Copy the Transform Scala below
    

    import java.sql.Timestamp
def transform(allClicks: Dataset[Row]) : Dataset[Row] = {
val ecommerceFullCatalog = spark.read.parquet("gs://training-ecommerce/catalog")
val someHardGoods = ecommerceFullCatalog.filter("department IN ('ACCESSORIES', 'APPLIANCE', 'COMPUTERS')")
val trimmedClicks = allClicks.select("query","doc_id","fusion_query_id","filters_s","type", "count_i","timestamp_tdt","user_id","id")
val hardGoodClicks = trimmedClicks.alias("TC").join(someHardGoods.withColumnRenamed("id", "doc_id"), Seq("doc_id")).select("TC.*", "name", "longDescription", "department")
val userClickCounts = hardGoodClicks.groupBy("user_id").count.withColumnRenamed("count", "user_count")
val itemClickCounts = hardGoodClicks.groupBy("doc_id").count.withColumnRenamed("count", "item_count")
val clicksWithCounts = hardGoodClicks.join(userClickCounts, Seq("user_id")).join(itemClickCounts, Seq("doc_id"))
val usefulClicks = clicksWithCounts.filter("user_count > 2 AND item_count > 4").drop("user_count","item_count")
val now = System.currentTimeMillis()
val maxDate = usefulClicks.agg(max("timestamp_tdt")).take(1)(0).getAs[Timestamp](0).getTime
val diff = now - maxDate
val addTime = udf((t: Timestamp, diff : Long) => new Timestamp(t.getTime + diff))
val newDF = usefulClicks
.withColumnRenamed("timestamp_tdt", "orig_timestamp_tdt")
.withColumn("timestamp_tdt", addTime($"orig_timestamp_tdt", lit(diff)))
newDF
}

    <script> function copyEvent(id) { var str = document.getElementById(id); window.getSelection().selectAllChildren(str); document.execCommand("Copy") }

  10. Close the Transform Scala text box. Click Save, then click Run, then click Start

    Note: Success! The job will take about up to three minutes, when it's done, the "running" icon will change to a "sunshine". If your job fails, be sure to check your parameters with the configurations above and run the job again.

Verify Successful Data Load

  1. Hover over QUERYING, click on Query Workbench

  2. Click Display Fields in the upper right corner and change the following fields:

  • Name field to name

  • Description field to longDescription

  1. Click Add a field facet and choose department

  2. Click Save

Note: If prompted, save over the existing Labs pipeline.

  1. Click the Collections dropdown and select Labs_signals

  1. Hover over QUERYING, click on Query Workbench

  2. Click Display Fields and change the Name field to name

  3. If there are no facets, click Add a field facet and choose department

  4. Click Save. Confirm that you will save over the existing pipeline.

    Note: Note that these signals contain all the data from the associated catalog item. This is NOT a recommended design for production, as the signals index would be very large. However, this can be very useful for our examples, as the click data is easier for us to read and comprehend.

  5. Make sure to scroll through all of your open Fusion tabs and Save each!

Great job! You now have a functioning Fusion app with data from utilizing the Parallel Bulk Loader! If you would like to save your Fusion App to reference later, you can do it now:

  1. Return to the Fusion Launcher
  2. Hover over your app and click on the cog that appears in the lower right corner
  3. Within the box that opens, click Export app to zip

This concludes the PBL Lab.

Start your environment by clicking Start Lab above.

The environment should begin to load immediately as indicated by the Vocareum loading symbol.Please do not click Start Lab again. It may take a few mintues for the Fusion environment to fully display.

When the Fusion Login page displays, login: * USERNAME: admin * PASSWORD: password123

In this lab you will inspect a collection, train a classifier model, implement classification at query time, and test the classifier pipeline! Let's get started by inspecting the collection on the app that you built in the first lab.

  1. Click on the app Labs to enter the Fusion workspace, this is where you can build and test the classifier model

    Inspecting the Collection

  2. Click the Collections dropdown in the top left navigation header and select Labs_signals

  3. Hover over the QUERYING icon in the sidebar, then click Query Workbench

    Note: In this lab, we will build a classifier that predicts which store department a user is interested in based on their input query. This involves finding correlations between certain query terms and the store department the user ultimately clicks into.

  4. Excecute the query:"washer dryer combo"

    Note: Everyone who searched for “washer dryer combo” also clicked on an item from the APPLIANCE department. Not every query is this clear-cut.

  5. Execute the query query:charger

    Note: These results are ambiguous, however there is still a clear trend towards ACCESSORIES. These patterns can be captured and learned by a Classifier; then used to influence relevancy at query time.

    Training a Classifier Model

  6. Hover over the COLLECTIONS icon in the sidebar, then click Jobs

  7. Click Add+, from the drop-down menu, select Logistic Regression Classifier Training

  8. Click Advanced. Let's add some parameters:

    • Change the Spark Job ID to department_qi_model
    • Update the Training Collection to Labs_signals
    • Update the Field to Vectorize to longDescription
    • Update the Label Field to department
    • Update the Auto-balance training classes to un-enabled
    • Change Grid Search with Cross Validation to enabled
    • Update the Elastic Net Weight to 0.2
Why these parameters?
Parameter Value Explanation
Spark Job ID department_qi_model Unique name for this job. This will also be Model name, so make it intuitive
Training Collection Labs_signals Collection from which to pull data.
Field to Vectorize longDescription The input field. Model will predict a label based on the contents of this field
Label Field department 'The output field. The model will write its prediction label here.
Auto-balance training classes un-enabled Ensure that all classes of training data have the same size
Grid Search with Cross Validation enabled Cross Validation is always enabled. This parameter also enables Grid Search, which will experimentally determine the “best” values for Elastic Net Weight and Regularization Weight
Elastic Net Weigh 0.2 The Elastic Net (link) allows smooth interpolation of the L1 and L2 regularization methods. There is no single “correct” value for this, but a small number between 0 and 1 is a good base.

9. Click **Save**, then click **Run**, then click **Start**

Note: Success! The job will take about a minute, when it's done, the "running" icon will change to a "sunshine".

Implementing Classification at Query Time

  1. Click the Collections dropdown and select Labs

  2. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  3. Click New in the upper right corner of the Query Workbench, then click Save in the upper right corner. In the Save Pipeline box, name the new pipeline Labs_department_QI, then click Save

  1. Click Add a Stage and select Machine Learning. Let's add some parameters:

    • Change the Label to department_QI_classifier

    • Update the Machine Learning Model ID to department_qi_model

    • Update the Model input transformation script to:

var modelInput = new java.util.HashMap(); modelInput.put("concatField", request.getFirstParam("q")); modelInput
  1. Update the Model output transformation script to:
request.putSingleParam("predicted_department", modelOutput.get("labelPredictedByFusionModel"));
  1. Click Apply, then click Cancel to close the query pipeline

Note: The classifier stage calculates a label in the predicted_department field of each query request but doesn’t do anything with that label. We will create another stage that uses the label to filter or boost documents based on the label.

  1. Click Add a Stage and select Additional Query Parameters. Let's add some parameters:

    • Change the Label to QI_department_filter

    • Under Parameters and Values, click + to add a new parameter

      • Change the Parameter Name to fq

      • Update the Parameter Value to department:"<predicted_department>"

  2. Click Apply, then click Cancel.

    • Using the three horizontol lines icon, drag the QI_department_filter stage so that it appears after the department_QI_Classifier stage in the Query Pipeline stages

  3. Click Save and save over the existing pipeline

Testing the Classifier Pipeline

  1. Using the Labs_department_QI pipeline, excecute the query ipod cover

  2. Click Add a field facet and select department

  3. Click Display Fields in the upper right corner of the Query Workbench and change the NAME display field to name

Note: The classifier associates the query (ipod cover) with COMPUTERS department. The department filter stage eliminates all documents from every other department.

  1. Click the green circle to disable the QI_department_filter query pipeline stage. Note that the department facet now includes documents from APPLIANCE, ACCESSORIES and COMPUTERS.

  2. Re-enable the QI_department_filters query pipeline stage and click on it to open the editor

Note: Filtering is not the only way we can utilize the classifier label. In fact, it is probably one of the least effective methods because, in cases where the classifier is “wrong”, the user-desired-department will not appear at all. Instead, lets use the classifier label to implement a relevance boost. In this case, the predicted department will still rise to the top of the results, but the other documents are still there to be found in case the prediction was wrong.

  1. Change the Parameter Name to bq and the Parameter Value to department:"<predicted_department>"^20

  2. Click Apply, then Cancel

Note: Now we get the same COMPUTERS documents in the top results as we did with a Filter Query, but without sacrificing the other documents in case of a classifier mistake.

  1. Make sure to Save your open Fusion Workspace tabs!

Great job! You now have a functioning query intent classification model! If you would like to save your Fusion App to reference later, you can do it now:

  1. Return to the Fusion Launcher
  2. Hover over your app and click on the cog that appears in the lower right corner
  3. Within the box that opens, click Export app to zip

This concludes the Building a Query Intent Classifier Lab.

Start your environment by clicking Start Lab above.

The environment should begin to load immediately as indicated by the Vocareum loading symbol. Please do not click Start Lab again. It may take a few mintues for the Fusion environment to fully display.

When the Fusion Login page displays, login:

  • USERNAME: admin
  • PASSWORD: password123

In this lab you will ingest data so that you can enhance relevancy using query rewriting! Let's get started by importing the relevancy app

Configuring the Relevancy App

  1. Click on the app relevancy-app to enter the Fusion workspace, this is where you can ingest the data

  2. Hover over the COLLECTIONS icon in the sidebar, then click Jobs

  3. Select and run each of the following jobs in the order listed below:

    • relevancy
    • relevancy_signals
    • relevancy_click_signals_aggregation

Note: Success! The jobs will take about up to three minutes each, when it's done, the "running" icon will change to a "sunshine". If the job fails initially, rerun the job again.

  1. Hover over the QUERYING icon in the sidebar, then click Query Workbench. Check the first few results to ensure the job was run successfully

Relevancy Enhancement Tactics

  1. Hover over the COLLECTIONS icon in the sidebar, then click Jobs. Click on and run relevancy_head_tail

Note: Success! The job will take about a few minutes, when it's done, the "running" icon will change to a "sunshine".

  1. Hover over the RELEVANCE icon in the sidebar, then click Rules. A new tab will open in your browser. Keep this tab open as we will revisit it through the training session.

  2. Click the Rewrite icon in the sidebar to access the Underperforming Query Rewriting dashboard

  3. Click Head/Tail and find dsl router in the list of rewrites

  1. Notice that Status is set to auto and that Published is set to yes. This is because the Head/Tail suggestion falls into the auto publish category set by the head-tail relevancy job we noted earlier in training.

    • Click on Auto and select Approved to change the status to approved.
    • The change in status will unpublish the rewrite. To publish, hover over the rule and click the green up arrow in the Published column. Confirm you want to publish by clicking Publish in the Publish Rules box that appears.

  2. Return to the Fusion tab. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  3. Toggle off the Text Tagger stage in the query pipeline by clicking on the green circle. Update the View As option to Debug. This is located in the bottom right side of the query workbench

  4. Execute a search for dsl router

Note: Notice what Fusion is using to perform this search in the rawquerystring and querystring sections

  1. Toggle on the Text Tagger stage by clicking on the grey circle. If the view does not automatically update, requery dsl router to see the change

Note: Notice the search enhancement when we use Query Rewrite principles by turning on the Text Tagger stage. This shows the search enhancement when we use Query Rewrite principles through the Text Tagger stage. Notice the updated search parameters Fusion is using to perform this search in the rawquerystring and querystring sections

  1. Hover over the COLLECTIONS icon in the sidebar, then click Jobs. Click on and run relevancy-spell-correction

Note: Success! The job will take about a few minutes, when it's done, the "running" icon will change to a "sunshine".

  1. Return to the Query Rewriting browser tab

  2. Click Misspelling and find any entry in the list of rewrites

  3. Change the Status to Approved and the Published to Yes. Remember which one you published becuase you'll need it in the next steps.

  4. Return to the Fusion tab. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  5. Change View As to Results

  6. Toggle off the Text Tagger stage by clicking on the green circle and execute a search for the misspelling rule you approved and published on the Query Rewrites page

  7. Toggle back on the Text Tagger stage by clicking on the grey circle and observe the difference in the relevancy of results

Note: Notice the search enhancement when we use Query Rewrite principles by turning on the Text Tagger stage. This shows the search enhancement when we use Query Rewrite principles through the Text Tagger stage. Notice the updated search parameters Fusion is using to perform this search in the rawquerystring and querystring sections

  1. Hover over the COLLECTIONS icon in the sidebar, then click Jobs. Click on and run relevancy-phrase-extraction

  2. Return to the Query Rewriting browser tab

  3. Click Phrase and find any entry in the list of rewrites

  4. Change the Status to Approved and the Published to Yes. Remember which one you published because you'll need it in the next steps.

  5. Return to the Fusion tab. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  6. Toggle off the Text Tagger stage by clicking on the green circle and execute a search for the entry you approved and published on the Query Rewrites page

  7. Toggle back on the Text Tagger stage by click on the grey circle and observe the difference in the relevancy of results

Note: Notice the search enhancement when we use Query Rewrite principles by turning on the Text Tagger stage. This shows the search enhancement when we use Query Rewrite principles through the Text Tagger stage. Notice the updated search parameters Fusion is using to perform this search in the rawquerystring and querystring sections

  1. Make sure to Save your open Fusion Workspace tabs!

Great job! You now have functioning query rewrites! If you would like to save your Fusion App to reference later, you can do it now:

  1. Return to the Fusion Launcher
  2. Hover over your app and click on the cog that appears in the lower right corner
  3. Within the box that opens, click Export app to zip

This concludes the Query Rewrites Relevancy Lab.

Start your environment by clicking Start Lab above.

The environment should begin to load immediately as indicated by the Vocareum loading symbol. Please do not click Start Lab again. It may take a few mintues for the Fusion environment to fully display.

When the Fusion Login page displays, login:

  • USERNAME: admin
  • PASSWORD: password123

In this lab you will set up clustering, evaluate results, and tune the clustering job! Let's get started by setting up clustering

  1. Click on the app Labs to enter the Fusion workspace, this is where you can build and test document clustering

Setting up Clustering

  1. Hover over the COLLECTIONS icon in the sidebar, then click Jobs

  2. Click Add, then select Document Clustering

  3. Toggle on Advanced. Let's add some parameters:

    • Change the Spark Job ID to product_clustering
    • Update the Training Collection to Labs
    • Update the Output Collection to Labs
    • Find Training Data Sampling Fraction and input 0.2
    • Update the Field to Vectorize to longDescription

  4. Click Save, then Run, then Start

Note: Success! The job will take about a minute, when it's done, the "running" icon will change to a "sunshine".

Evaluating Results

  1. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  2. Click Add a field facet and select cluster_label

  3. Click Add a field facet and select freq_terms

  4. Click View next 10 under the cluster_label facet

Note: The cluster_label field contains the five terms that best define the center of each cluster. The freq_terms field contains the five terms that appear most often in each cluster. You will note that, in many cases, the frequent terms in a cluster are also among those those that best define it. The clustering job also identified and labeled oddball documents: abnormally long ones, those too short to determine anything meaningful, and those that do not seem closely related to any of the big clusters.

Tuning the Clustering Job

  1. Hover over the COLLECTIONS icon in the sidebar, then click Collections Manager

  2. Click New, and name your new collection Labs2.

  3. Toggle on Advanced and uncheck Enable Signals. Click Save Collection

  4. Hover over the COLLECTIONS icon in the sidebar, then click Jobs

  5. Select the product_clustering job and toggle on Advanced

  6. Modify the following parameters:

    • Change the Output Collection to Labs2
    • Update the Outlier Cutoff to 0.02
    • Update the Min Possible Number of Clusters to 5
    • Update the Max Possible Number of Clusters to 25
    • Update the Word2Vec Dimension to 100

  7. Click Save, then Run, then Start

Note: Success! The job will take about a minute, when it's done, the "running" icon will change to a "sunshine".

  1. Click the Collections dropdown and select the output collection Labs2

  2. Hover over the QUERYING icon in the sidebar, then click Query Workbench

  3. Click Add a field facet and select cluster_label

  4. Click Add a field facet and select freq_terms

Note: Explore your clusters. Do they look better than they did previously?

  1. Make sure to Save your open Fusion Workspace tabs!

Great job! You now have functioning document clustering in your Fusion instance! If you would like to save your Fusion App to reference later, you can do it now:

  1. Return to the Fusion Launcher
  2. Hover over your app and click on the cog that appears in the lower right corner
  3. Within the box that opens, click Export app to zip

This concludes the Document Clustering Lab.

Start your environment by clicking Start Lab above.

The environment should begin to load immediately as indicated by the Vocareum loading symbol. Please do not click Start Lab again. It may take a few mintues for the Fusion environment to fully display.

When the Fusion Login page displays, login:

  • USERNAME: admin
  • PASSWORD: password123

In this lab you will configure recommendations, verify successful model builds, test item-item recommendations and test user-item recommendations! Let's get started by enabling recommendations

  1. Click on the app Labs to enter the Fusion workspace, this is where you can build and configure recommenders.

  2. Hover over the RELEVANCE icon in the sidebar, then click Recommendations.

  3. After reading the dialog box, click Enable Recommendations.

Configuring Recommendations

  1. Hover over the COLLECTIONS icon in the sidebar, then click Jobs.

  2. Click Labs_item_recs, then toggle on Advanced.

Note: There are two main tasks involved in setting up items-for-user recommendations. First, it uses existing signals to determine each past user’s level of interest in each item. Second, it uses past user-item preferences to create a user-item matrix, which can be used both to recommend items to new users and recommend other items that tend to collocate with the one currently clicked.

  1. Let's assign job parameters:

    • Expand TRAINING DATA SETTINGS and update the Training Data Sampling Fraction to 0.05
    • Expand ITEM METADATA SETTINGS and update the Item Metadata Collection to Labs
    • Under ITEM METADATA SETTINGS, update the Item Metadata Join Field to id
    • Under ITEM METADATA SETTINGS, locate the Add Item Metadata Fields and click add to department in the text field that appears

  2. Click Save, but do not run the job yet

The Labs_user_item_prefs_agg job will calculate user-item preferences based on past behavior. The output will be used by Labs_item_recs to create a user-item matrix. This is why we configured that job first.

  1. Click Labs_user_item_prefs_agg and toggle on Advanced.

  2. Uncheck Aggregate New and Merge with Existing, then click Save. These parameters are only functional for subsequent runs of the job, not on the first run.

  1. Click Run, then click Start.

Note: Success! The job will take about a minute, when it's done, the "running" icon will change to a "sunshine".

  1. Return to the Labs_item_recs job. Click Run, then click Start.

Note: Success! The job will take about a minute, when it's done, the "running" icon will change to a "sunshine".

Verify Successful Model Build

  1. Click the Collections dropdown and select Labs_items_for_user_recommendations.

  2. Hover over the QUERYING icon in the sidebar, then click Query Workbench.

  3. Click show fields under any document.

Note: The key among the fields here are itemID and userID, representing a single, unique user-item pairing. The weight field is a prediction of how well this item matches this user, based on the behavior of other users with similar activity. Also note the department field. This was pulled in by the metadata JOIN we requested in the Labs_item_recommendations job. This field can be used at query time; for example, to only recommend items from certain departments.

  1. Click the Collections dropdown and select Labs_items_for_item_recommendations.

  2. Hover over the QUERYING icon in the sidebar, then click Query Workbench.

  3. Click show fields under any document.

Note: Notable fields here are itemId and otherItemId representing a single unique pairing of items. The sim field is a measurement of how “similar” these two items are based on collocation in user activities. Also note the department field from the metadata JOIN.

Testing Item-Item Recommendations

  1. Click the Collections dropdown and select Labs.

  2. Hover over the QUERYING icon in the sidebar, then click Query Workbench.

  3. Click Load in the upper right corner of the screen and select Labs_items_for_item_recommmendations.

  4. Select the Recommend Items for Item stage. Let's fill out the following parameters:

    • Set the Boost Param to bq

Note: This stage implements a relevance boost for similar items when a single item is “selected” via an item_id parameter. This parameter will be populated, for example, by going to the details page for a given item. The last three parameters: Item ID Field, Recommended Item ID Field, Similarity Score Field relate to the structure of the items-for-item documents examined previously

  1. Click Apply, then click Cancel to close the stage.

  2. Click Parameters in the upper right side of the Query Workbench and select Edit parameters.

  3. Click the green + and add a parameter with a Parameter Name of item_id and a Value of 2460117.

  4. Click Close to apply the parameter.

Note: This parameter prompts the Recommend Items for Item stage to boost items similar to the one with doc id 246011.

  1. Execute a query for id:2460117 and click show fields.

Note: This item is an SSD drive. The recommender boosts items related to hard drives.

  1. Execute a query for *:*, and note that items similar are boosted.

  2. Switch to Labs_items_for_item_recommendations collection.

    • If prompted, abandon changes to the existing pipeline
    • Navigate to the Query Workbench and execute a query for sim:[.997 TO *], and add a field facet of itemId. Pick out any one itemId in the facet list and remember the itemId value for Steps 29-31.

Note: Any items with 10 similar items at such high similarities should look pretty good. Feel free to adjust the lower bound of the sim query to find the best candidates.

  1. Switch back to the collection Labs.

  2. Replace the existing item_id parameter with the noted item_id parameter (from Step 28) in the top right of the query workbench.

  3. Disable the Recommended Items for Item stage in the query pipeline panel and note how the items are back to a pseudo-random document ordering. Re-enable Recommended Items for Items

Testing User-Item Recommendations

  1. Click Load in the upper right corner of the Query Workbench and select Labs_items_for_user_recommendations

Note: There are two ways to implement a user-item recommendation. First is personalized recommendations where we offer users a list of interesting items before they ask for anything. Second is personalized Search where we apply boost factors to items based on the user-item similarity of the person looking.

  1. Select the Recommend Items for User stage.

Note: The key query parameter is user_id that will generally be populated by a login action The user-item recommendations collections will be searched for items that have a userId field equal to user_id, and will provide a relevance boost on the associated itemId. The weight determines the magnitude of the boost.

  1. Click Cancel to close the stage. Click Display Fields and change Name to name. Change Description to longDescription.

  2. Click Parameters, then select Edit Parameters.

  3. Delete the item_id parameter and add a new parameter with the Parameter Name of user_id and a Value of 8423cdfb8a15894e414ca32c21f68a95aacc0326. Click Close.

Note: Enabled parameters should have a green circle next to the parameter name. You can observe the difference in results when the Recommend Items for User stage is disabled by clicking the green circle.

  1. Execute a query for *:*, and note the items listed.
  1. Execute a query for laptop and note the difference in item order.

Note: The top result for this user was originally 5 ranks lower, without the recommender active. This is an example of Personalized Search. For better results, use a large sample when building the recommender model.

  1. Make sure to Save your open Fusion Workspace tabs!

Great job! You now have a functioning recommender! If you would like to save your Fusion App to reference later, you can do it now:

  1. Return to the Fusion Launcher
  2. Hover over your app and click on the cog that appears in the lower right corner
  3. Within the box that opens, click Export app to zip

This concludes the Building a Recommender Lab.