Questions tagged with Amazon Personalize

Using a web console for Amazon Personalize for creating item datasets, it throws an error "GENRES field cannot be of type null" when trying to create an item schema with GENRES that contains type "null". But the documentation https://docs.aws.amazon.com/personalize/latest/dg/how-it-works-dataset-schema.html has GENRES with type null for imperfect data. Also, it throws an error "schema is missing required field [genres]" if it's not included in the schema, which is odd, because the documentation says that the item schema requires ITEM_ID, CREATION_TIMESTAMP, and 1 metadata field where the table doesn't even mention genres. https://docs.aws.amazon.com/personalize/latest/dg/custom-datasets-and-schemas.html#dataset-requirements However, we could create the schema without genres when using the AWS SDK. Is this a bug in AWS console? or is it a new requirement that hasn't been updated on SDK?

Assume a short video recommendation system with a million short videos and users. The users can watch & like videos. Using a receipt such as USER_PERSONALIZATION (Users, Video as items, and Watched/Liked as interactions) the model will include thousands of interactions per user. To filter out the watched videos before recommending videos to the user, we can use Amazon Filter with: EXCLUDE ItemID WHERE Interactions.EVENT_TYPE IN ("watched"). However according to https://docs.aws.amazon.com/personalize/latest/dg/frequently-asked-questions.html "...when filtering, Amazon Personalize considers up to 200 historical interactions for a user, and up to 100 streamed interactions". Therefore, for such use case, we might still recommend watched videos, as there will be more than 200 watched videos per user and the filter only considers up to 200 watched video (interactions). Does it mean that Amazon Personalize is not considered a suitable solution for short video apps?

After created a Batch inference job to get Batch Recommendation I cannot figure it out how to delivery recommendation to my application (React Native) from file stored on S3 Bucket.

I was testing AWS Personalize recommenders and I realize that I was using all which charge for real-time recommendations. Is there a way to use them not for charge on real-time?

I am considering implementing AWS Personalize recommendations on a news site, to recommend articles that are relevant to the user. But scalability is important to us, so today we make heavy use of CloudFront to deliver cached pages. That means all users get the same page, with the same recommendations. How can I implement personalized recommendations without ruining this cachability? **Does AWS Personalize offer anything similar to segmentation, as described in the below article?** I.e. categorize a user into one of 10-or-so groups (preferably automatically), and then deliver recommendations that are relevant to that group as a whole, rather than that one user. https://wpengine.com/support/personalization-user-segmentation-page-cache/ (Not to be confused with AWS Personalize's user segmentation recipes, which seem to be a different thing.)

Say, we have some users, items in an online shop. We also keep their interaction data. Now say, metadata(Categorical list) of a user had a little update i.e. some data is added to the list. So if we then do a full training, the training will lead to incorrect recommendations. Because the interactions of that user were when the user had the old data. Now one may say we need to delete those interactions and the problem is solved. But in that case, we are losing interaction which is not good. The metadata(Categorical list) is updated i.e. some data is added not replaced. So delete is not an option. Let me give an example. In a shop, each item/product has one or more prerequisite licenses. That means only the users who have those licenses can buy and use those items. The user collects the licenses from the other place which is not in the concern of the shop. They collect licenses when they need them. User Schema: ``` { "type": "record", "name": "Users", "namespace": "com.amazonaws.personalize.schema", "fields": [ { "name": "USER_ID", "type": "string" }, { "name": "LICENSES", //Currently available licenses of the user "type": "string", "categorical": true }, ], "version": "1.0" } ``` Item Schema: ``` { "type": "record", "name": "Items", "namespace": "com.amazonaws.personalize.schema", "fields": [ { "name": "ITEM_ID", "type": "string" }, { "name": "PREREQUISITE_LICENSES",//The licenses user need to have to buy and use the item "type": "string", "categorical": true } ], "version": "1.0" } ``` Now say a user has license L1 and L2 and he has bought and used some products which needed license L1, L2, or L1 & L2. We have all the interactions. Now if that user bought another License L3, and we need to a full training that time, it will seem like the user with license L1, L2, L3 has only interacted with those items(L1, L2 items). But in reality, the user interacted with those(L1, L2) items because at that time he had not the license L3. It will give wrong information to the model. This can happen for other users too. In that case, the recommendation will have lower accuracy. How to handle that situation? Is there any Idea?

Hi, I am trying to train an Amazon Personalize Item recommendation solution using User Personalization recipe. I am trying to set the BPTT value explicitly to 6 and hidden dimension to 133. When I say create solution, these parameters are getting reset to 32 and 149. If I make the perfrom HPO as TRUE, these values are not defaulting to 32 and 149, different value is getting suggested.

I've seen a situation (in retail) where Amazon Personalize [re-ranking](https://docs.aws.amazon.com/personalize/latest/dg/native-recipe-search.html) and [user-personalization](https://docs.aws.amazon.com/personalize/latest/dg/native-recipe-new-item-USER_PERSONALIZATION.html) solutions seem to be over-indexing on recommending the most popular items in the catalogue, even for users with different tastes and a decent amount of interaction history (results are not exactly the same, but popular items seem very heavily weighted). These recipes don't seem to have the `popularity_discount_factor` hyperparameter like [SIMS does](https://docs.aws.amazon.com/personalize/latest/dg/native-recipe-sims.html)... So are there any particular suggestions for reducing or avoiding this effect?

Hey guys, I have a question about the Amazon Personalize Recipe, customers-who-viewed-x-also-viewed. I tried to call this recipe, but I got the error. It's like this: "InvalidInputException: An error occurred (InvalidInputException) when calling the CreateSolution operation: This recipe with ARN arn:aws:personalize:::recipe/aws-ecomm-customers-who-viewed-x-also-viewed is unsupported. Please use the supported recipes listed at https://docs.aws.amazon.com/personalize/latest/dg/working-with-predefined-recipes.html" I wanna know how to call the recipe. Thank you for your help.

Hi guys, I am using the user-personalization recipe in Amazon Personalize to build a recommendation system. I am wondering if there is any way to download the recommended lists of users as bulk, and also wondering if I can get the result of true class(if it's truly recommended for the user). Cuz, there is only value indicators like ndcg and precision. However, I wanna see through my eyes what is happening inside of the black box(user-personalization). Thank you.

I've explored amazon personalize, etc for generating recommendations. Amazon personalize can be used when all the content is with the company/a single entity. For example, in Netflix, all the content (the catalogue of movies, tv shows etc.) is with them and they generate personalized movie/tv show recommendations. But what if there's a platform similar to Youtube, TikTok, where users can: - post content (users are continuously generating content) - view other users content and interact(like, share, repost, comment) - follow other users When there is user generated content like this and users follow other users (meaning they probably want recommendations from users they follow), how do we give recommendations? Can we do it with Amazon Personalize? What algorithms and tools can be used? **Lots of content - handling the cold start problem** And when there is user generated content, there is going to be lots of content being generated every minute. So how do we handle the cold start problem (i.e. how do we decide who to recommend all of this new influx of content too)? Usually we might experiment with this new content, like recommend it to some users , see how they're responding and appropriately decide how to recommend this content. But when there is a very high frequency of content being created, how do we reduce the amount of time it takes to give recommendations/push the new content to users quickly? And does anybody know if the questions mentioned above can be addressed using Amazon Personalize (in any way)? Open to any and all suggestions. Thank you!

We use domain dataset group at Amazon Personalize. We finished to import data (interactions, users, items) and create an recommender (Recommended for you). We can get recommendations from the recommender with the Amazon Personalize console. Then, we would like to get recommendations with the SDK for Python (Boto3). We sent request per following Developer Guide, however ParamValidationError occured. Getting recommendations with a recommender (AWS SDKs): https://docs.aws.amazon.com/personalize/latest/dg/domain-dsg-recommendations.html#get-domain-rec-sdk How can we get recommendations from Domain dataset recommender with the SDK for Python (Boto3)? We tried following pattens; **pattern1** Request ``` response = personalizeRt.get_recommendations( recommenderArn = 'Recommender ARN', userId = 'User ID', numResults = 10 ) ``` Error ``` ParamValidationError: Parameter validation failed: Missing required parameter in input: "campaignArn" Unknown parameter in input: "recommenderArn", must be one of: campaignArn, itemId, userId, numResults, context, filterArn, filterValues ``` **pattern2** Request ``` response = personalizeRt.get_recommendations( campaignArn = 'Recommender ARN', userId = 'User ID', numResults = 10 ) ``` Error ``` InvalidInputException: An error occurred (InvalidInputException) when calling the GetRecommendations operation: The given campaign ARN is invalid: Recommender ARN ```

1) For the VIDEO_ON_DEMAND domain, some use cases include multiple event types. For example, the 'Top picks for you' use case includes two event types 'watch' and 'click'. Is 'watch' given more weight than 'click' when training the model? In general, when there is more than one event type, do domain recommenders give more weight to some event types? 2) In our use case, we have a platform that recommends video content. However, we have multiple event types, and some events need to be given more weight than others. Below is the list of our event types in the order of their importance: SHARE > LIKE > WATCH_COMPLETE > WATCH_PARTIAL > STARTED > SKIP So when training the model, we would want 'SHARE' to have more weight than 'LIKE', and 'LIKE' to have more weight than 'WATCH_COMPLETE' and so on. I was looking into custom solutions. It looks like there is no way to give weights when using Personalize's custom solutions as mentioned in this [post](https://stackoverflow.com/questions/69456739/any-way-to-tell-aws-personalize-that-some-interactions-count-more-than-others/69483117#69483117)... --- **So when using Amazon Personalize, should we use domain recommenders or build custom solutions for our use case?** **If we cannot give weights to different event types using Personalize, then what are alternatives? **Should we use Amazon SageMaker and build models from scratch? *Open to any and all suggestions.* Thank you!

In the documentation, on the [quotas page](https://docs.aws.amazon.com/personalize/latest/dg/limits.html#limits-table) there is information about the minimum and maximum limits for the datasets we give to personalize. But can someone suggest how much interaction data you would need in order to generate decent/good recommendations? I used the [movielens-100k dataset](https://grouplens.org/datasets/movielens/100k/) and set `performHPO` to `True`. For the trained solution, the normalized discounted cumulative gain @ 25 was 0.2977 and the precision @ 25 was 0.0512. Is this low performance due to not having enough interaction data?

When configuring a solution, how do we specify that we want to include multiple event types during training? In our dataset, we have three event types "STARTED", "PARTIALLYWATCHED", "COMPLETED". We only want to include "PARTIALLYWATCHED" and "COMPLETED" during training. In the console, we tried giving it in the following ways: - `PARTIALLYWATCHED,COMPLETED` - `PARTIALLYWATCHED|COMPLETED` But the solution version failed and we got the following error: **"Dataset has fewer than 1000 interactions after filtering by event type: PARTIAL|COMPLETED". ** However, in our interactions dataset, we have around 7000 interactions with event types "PARTIALLYWATCHED" and "COMPLETED". **How do we specify that we want to include multiple event types during training when creating a solution?**

In the [documentation](https://docs.aws.amazon.com/personalize/latest/dg/API_CreateSolution.html#personalize-CreateSolution-request-performAutoML) it says that when `performAutoML` is set to `true`, amazon personalize analyzes your training data and **selects the optimal USER_PERSONALIZATION recipe** and hyperparameters. In this case, **you must omit recipeArn.** 1. What does it mean by selecting the optimal USER_PERSONALIZATION recipe..are there multiple USER_PERSONALIZATION recipes? When looking at the documentation, it seemed like there was only one and we would use it's recipeArn while creating a solution. 2. It also says AutoML lengthens the training process as compared to selecting a specific recipe. So what's the cost difference (approximately) if we set it to False vs True? Is it very significant?

When we use [CreateSolution](https://docs.aws.amazon.com/personalize/latest/dg/API_CreateSolution.html), what is the difference between **algorithmHyperParameters** and **algorithmHyperParameterRanges**? In the documentation, it says that in [algorithmHyperParameterRanges](https://docs.aws.amazon.com/personalize/latest/dg/API_HPOConfig.html#personalize-Type-HPOConfig-algorithmHyperParameterRanges) we should specify the hyperparameters and their allowable ranges and for [algorithmHyperParameters](https://docs.aws.amazon.com/personalize/latest/dg/API_SolutionConfig.html#personalize-Type-SolutionConfig-algorithmHyperParameters) we should list the hyperparameter names and ranges. How are they different and when do we use what? Can you please provide an example? Thank you!

I'm exploring using the new [CloudWatch Evidently](https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitoring/CloudWatch-Evidently.html) feature for measuring success of recommendation model deployments with [Amazon Personalize](https://docs.aws.amazon.com/personalize/latest/dg/what-is-personalize.html). In this context, assigning a user or session to a particular feature variation (baseline recommendations list vs Personalize campaign 1 vs Personalize campaign 2) **might** trigger one *or more* valuable "events": - Maybe a click/view for an individual item (What's this item's price? How far down the recommendation list was it?) - Maybe a checkout for a basket of products (with an overall total price or total margin) If I understand right (?), Evidently experiment dashboards for "statistical significance" and "improvement" today look just at the **distribution of values** of recorded events in terms of averages and distribution, right? The number of data points is used for assessing "how significant" but not "what's better"? If so, this seems like a challenge for "optional" events: For example what if one treatment gives me really high basket value on average (only recommending expensive products), but very few users convert? I could see really high metrics for the new treatment, even though its overall value was very poor. Do I understand correctly here? And if so, how might you recommend defining Evidently metrics for these kinds of use cases? For example maybe we'd need to find a way of generating zero-value metric events when a session is abandoned?

I did the amazon personalize deep dive series on youtube. At the timestamp [8:33 in the video](https://youtu.be/TEioktJD1GE?t=513), it was mentioned that **'Personalize does not understand negative feedback.' and that any interaction you submit is assumed to be a positive one.** But I think that giving negative feedback could improve the recommendations that we give on a whole. Personalize knowing that a user does not like a given item 'A' would help ensure that it does not recommend items similar to 'A' in the future. **Is there any way in which we can give negative feedback**(ex. user doesn't like items x,y,z) **to amazon personalize**? --- A possible way to give negative feedback that I thought of: Let's say users can give ratings out of 5 for movies. Every time a user gives a rating >= 3 in the interactions dataset, we add an additional interaction in the dataset (i.e we have two interactions saying that a user rated a movie >=3 in the interactions.csv instead of just one). However, if he gives a rating <=2 (meaning he probably doesn't like the movie), we just keep the single interaction of that in the interactions dataset (i.e. we only have one interaction saying that a user rated a movie <=2 in the interactions.csv file) Would this in any way help to convey to personalize that ratings <=2 are not as important/that the user did not like them?

I'm trying to get Personalize to work with my data, and it looks like all requirements are met, but I'm still getting "InternalServerError: We encountered an internal error. Please try again." error when initializing Solution creation. Batch imports of Users, Items, and Interactions went well, so it checks out with schemas so I don't have any idea on what's wrong here. Is there a way to get more specific error message or an example of invalid data?

From what I can tell the standard use case for Personalize is to get Item recommendations for a User. Is there a way to get User recommendations for an Item? My use case is sending out a mass email about a specific item going on sale. I only want to send the emails to customers who may be interested in that item. Ideally i'd be able to submit a job to Personalize to give me a list of all Users with greater than n score for this Item. Is this possible with Personalize? Thanks!

How quickly do events inserted with putEvents influence recommendations for the user? Will the recommendations be instant, or is there any pre-defined interval where the new event data gets processed to create new recommendations?

I have been trying to delete the user-item interactions dataset, but the data imported with putevents(recorded events with event tracker) still remain, only bulk data gets deleted Any solutions?

How to generate a solution version(USER_PERSONALIZATION recipes) using only recent records when importing records incrementally? I’m importing records(interactions, items) incrementally since new items are created every time. Since importing records incrementally makes training time(price) very high and the integrated records contains unnecessary old items, I want to generate a new solution version using only recent records(interaction, items). Is there any ways to do this? As far as i know, there is no way to remove old records when generating a new solution version. Currently, i’m generating a new solution version on all items(old and new) and remove the old items using filter when recommending items. **Q1. If I want to get a new solution version using only recent records, Do I have to create a new DatasetGroup?** **Q2. If new items are created every time and old items are unnecessary(e.g. news article recommendations), what is a proper personalize architecture?** (The "architecture" means a pipeline managing recent dataset) Thanks for reading. Edited by: hdseo on Aug 1, 2021 12:09 AM Edited by: hdseo on Aug 1, 2021 12:22 AM Edited by: hdseo on Aug 1, 2021 12:26 AM

Is the sessionId field in PutEvent related to the session in the Session-based HRNN model? Is it used only for associating events with a user before they log in?

> InternalServerError: We encountered an internal error. Please try again. I've tried to recreate the import job but fails with the same error every time. Example job name: "item-_1e105c531-4d00-437d-b260-6ea789ca1c00". How can I get more visibility into what went wrong? What's the underlying problem for this error? Thanks,

Are there any tools or samples showing how to perform A/B testing or multi-arm bandit testing for Amazon Personalize solutions?

I'm considering using the `User-Personalization` recipe with the new [PutItems](https://docs.aws.amazon.com/personalize/latest/dg/API_UBS_PutItems.html) and [PutUsers](https://docs.aws.amazon.com/personalize/latest/dg/API_UBS_PutUsers.html) APIs, but with requirements on making sure that **new users whose metadata is known** receive only appropriate recommendations. For example after initial sign-up a user's country/age/membership status/etc are known and may be posted to Personalize `PutUsers`... As I understand the filtering mechanism is quite separate from the personalization model, but then I think in the past I heard of some differences between how new users and new items get picked up in modelling. Will new users' attributes be reflected in recommendation filters straight away (after the typical <15min near-real-time delay as for new/updated items?) Or would new users continue to receive out-of-filter recommendations until the next model re-train? Maybe even until the next `FULL` model re-train, if an [UPDATE training](https://docs.aws.amazon.com/personalize/latest/dg/API_CreateSolutionVersion.html) is not sufficient to onboard new users?

What is the minimum interaction needed for an item so that Amazon Personalize SIMS return a custom SIMS list instead of top popularity? What happens if a product has only 1,2,3 historical interactions?...

I ran into a case where Personalize SIMS recipe returns as top similar items the same list for almost every product, which are pretty much the top popular items. How to avoid Personalize doing that? I see there is a popularity_discount_factor parameter but its use is unclear. Happy to be clarified and to hear real stories of this being fixed.

I am trying to train a user-personalization solution in AWS personalize. I am exclusively using real time events i.e. I have not imported any users, items and interactions. I have created an event tracker and pumped more than 1000 events with more than 2 events for 25 users (this limit is mentioned in the documentation). However with just the real time interactions ingested by invoking the "put_events" API, the model training fails with the following message : "An error occurred (InvalidInputException) when calling the CreateSolution operation: Dataset group must contain an INTERACTIONS dataset with successfully completed dataset import job or EVENT_INTERACTIONS" Not having an INTERACTIONS dataset with successfully completed dataset import job, I understand and do not have. But what is this EVENT_INTERACTIONS ? I don't see anything mentioned anywhere. Finally, I have to ask, is it possible to train a solution using a user personalization recipe (aws-user-personalization) with only real time events and no other data ? Will appreciate any insights. P.S. This is implemented in python. Edited by: CarcharodonCarcharias on Oct 26, 2020 1:13 AM

Is Amazon Personalize relevant for non-session based activities ? Let's say for example that I have a dataset consisting of a list of products used by customers. Eg - customer 1 consumes product A, B, C - customer 2 consumes product A and C - etc This does not represent sessions. Is **Amazon Personalize HRNN** still relevant for this?

Hi, To update item data, I replaced item csv in S3 and created a new dataset import job via API. But there is no API to delete old import job. Is there any way to do it?

I have an interaction dataset that contains the following attributes: **USER_ID**, **ITEM_ID**, **TIMESTAMP**, **COUNTRY**, and **DEVICE**. I also have an event collection policy that collects the following attributes: **USER_ID**, **ITEM_ID**, **SESSION_ID**, **EVENT_TYPE**, **COUNTRY**, **DEVICE**, and **WIDGET**. Is it possible to create an Amazon Personalize model using a smaller combination of these attributes only? For example: **USER_ID**, **ITEM_ID**, **TIMESTAMP**, and **DEVICE**.

[This blog post][1] about "contextual recommendations" says this feature requires a "context aware solution with a User Personalization or Personalized Ranking recipe". In the blog, hrnn-metadata is not listed as a compatible recipe yet it is used it in the demo afterwards. **What is a context aware solution? Which Personalize recipe can learn from interaction context?** [1]: https://aws.amazon.com/blogs/machine-learning/increasing-the-relevance-of-your-amazon-personalize-recommendations-by-leveraging-contextual-information/

I created a Personalize dataset group with an item and an interaction dataset. I also activated the data collection. Is there any way to verify that things are working fine and to see the imported and collected data?

A customer is looking to make video recommendations using a small data set of just 150 videos including metadata on each video. They currently do not have any data on user interaction with the videos at this point as they haven’t released them yet to their customers. - Are they still able to utilize Personalize to begin making initial recommendations or do they need this interaction data (1,000 interactions/25 customers/2 interaction entries)? - For example - if a user finishes watching a video, could they recommend the next video based on similar genres/meta data of the videos themselves? And then of course integrate the user interaction data that they generate over time to make stronger recommendations. Ultimately, I'm wondering if this lack of user interaction data is a blocker to utilizing Personalize or if it simply will result in weaker recommendations. If it's a blocker to Personalize, is there a better solution for basic recommendations for a team without any ML experience?