[Employee Center] The "Recommended for You" widget is displaying the same "standard" KB articles and Catalog Items for some users, but for other users, it returns more personalised items in this widget that are relevant to the userIssue The Recommended for you widget calls the Script Include - RelevantForYouUtilSNC. This checks to see if the system property [sn_hr_sp.use_similar_user_query] is set to "true", and calls the Script Include - sn_hr_sp.HR_MLPortalUtilsSNC. This script include is responsible for getting the Similar HR Profiles based on the user ID, and has the following comment - /* getSimilarProfiles:Gets the similar HR profiles of the logged in user.The first time when user is logged in , a call is made to ML prediction server to fetch the similar profiles, these results are stored in ml_predictor_results table. Hence when the user is logged in from second time onwards results are fetched from ml_predictor_results tableParameters: 1) userId: userId of the logged in user2) topN: gets the topN similar HR Profiles*/ It gets the Similarity solution name from the Employee Profile AI Configuration, and the top 25 predicted results for similar HR Profiles is stored in the [ml_predictor_results] table, as per the topN option in the Script Include - sn_hr_sp.HR_MLPortalUtilsSNC, as follows - The out-of-box Similarity solution is User profile based recommendation, as this has been specifically designed to get similar HR Profiles for the logged in user that will display the recommendations based on the Input fields. Of course, you can create a custom solution and add/remove additional fields, but we would not recommend using table [sys_user] for this solution, as this feature is designed to use HR Profiles in table [sn_hr_core_profile]. This out-of-box Similarity solution has the following Input fields - user.department.nameposition.positionuser.manager.nameuser.manager.manager.nameuser.manager.manager.manager.nameuser.country The idea here is that User.Manager.Name and User.Manager.Manager.Name will look at similar Knowledge Articles and Catalog Items as the people they manage. Same with users from the same department or country, and who have the same position.The Script Include sn_hr_sp.HR_MLPortalUtilsSNC then calls UserRecommendationUtilSNC. This is the main script include that retrieves the KB articles, Catalog Items and Record Producers based on certain criteria discussed in the Resolution section.CauseFor users, when the prediction returns similar users who haven't got any data in tables [kb_use] or [sc_req_item] or [sc_item_produced_record] in the last 30 days, it will fall back and display the most viewed articles (all-time) or the most Requested Catalogs / most Requested Record Producers in the last 30 days.ResolutionQ1. Can you confirm as to why certain users aren't seeing more specific items in the "Recommended for you" widget. Is it to do with the the lack of Knowledge Article and Catalog Item "usage" data for these users? A. Yes, Certain users might not be getting similar users for them, or if they do, there might not be much usage from those users to fulfil the "limit" and "recentCutoffDate" mentioned in the instance options on the "Recommended for You" widget. There are 3 tables used in case of similar users found to determine the similar items to be displayed. Let's suppose the limit is 6 and recentCutOffdate is 30 days - a. Table [kb_use]: Similar users who used the KB articles in the last 30 days are consideredb. Table [sc_req_item]: Similar users who requested items in the last 30 days are consideredc. Table [sc_item_produced_record]: Similar users who requested record producer in the last 30 days are considered We pick items as per the limit for KBs and Catalog Items. If not being fulfilled, then empty spaces are filled with the most viewed articles (all-time) or the most Requested Catalogs / most Requested Record Producers in the last 30 days. So, users who may not have similar users in the Similarity solution prediction result or there is no "usage" of similar users, then this will lead to displaying the generic most popular items. Q2. If it is a lack of data, can this threshold be confirmed and is there a mechanism to change this? A. You can manipulate the trained Similarity model threshold to a lower value, so you would get more similar users in most cases. You can also increase the recentCutOffDate, but this will also impact performance directly proportional to how far the last number of days you want to consider. We do not recommend increasing this to a very high number, as the time factor should also consider the popularity of an item amongst their peers, it also an intensive process based on grouping and finding similar users. Q3. When looking at this "usage" data what time period is it looking at, e.g. last 6 months or forever? A. It is based on the recentCutOffDate in the instance option of the "Recommended for You" widget that determines which Knowledge Articles (sys_view_count of kb_articles) and Catalog Items/Producers (most ordered in the last recentCutOffDates)