Recommendation

Recommendation systems with Collaborative Filtering help suggest relevant items by analyzing the behavior of different users or similar items. There are two main approaches: user-user and item-item.

For example, imagine you liked a movie on a streaming platform. With the user-user method, the system looks for people with similar tastes to yours and suggests movies they enjoyed. On the other hand, with the item-item method, it may suggest a movie based on what you just watched.

These methods work like an "invisible network of friends," helping you discover new things that perfectly match your interests as a user.

Currently, only the Collaborative Filtering model is available, but other models will be developed soon.

EXAMPLE OF USE

uses
  URecommender;
  
procedure ExampleRecommender;
var
  vRecommender: TRecommender;
begin
  vRecommender := TRecommender.Create('C:\DelphAI\DelphAI\Datasets\MovieLens-100k\user_item_matrix.csv', 5, 20, TUserScoreAggregationMethod.amWeightedAverage, dmCosine, True, False);
  try
    ShowArray(vRecommender.RecommendFromUser(105));
    ShowArray(vRecommender.RecommendFromItem(4070));
  finally
    vRecommender.Free;
  end;
end;

CLASSES AND METHODS GUIDE

TRecommender:

• FDataset : TAIDatasetClassification; : access the current database contained in the model.

• procedure ClearDataset; : cleans the dataset.

• constructor Create(aMatrixFile : String; aItemsToRecommendCount, aK : Integer; aAggregMethod : TUserScoreAggregationMethod = amWeightedAverage; aDistanceMethod : TDistanceMode = dmCosine; aCalculateItemDistanceOnCreate : Boolean = False; aHasHeader : Boolean = True); or

• constructor Create(aMatrix: TDataSet; aItemsToRecommendCount, aK : Integer; aAggregMethod : TUserScoreAggregationMethod = amWeightedAverage; aDistanceMethod : TDistanceMode = dmCosine; aCalculateItemDistanceOnCreate : Boolean = False); : creates the object with the dataset that will be used for training.

  • aMatrixFile : CSV file path.

  • aMatrix : TDataSet object that contains the data that will be used for training. See what the structure should look like here.

  • aItemsToRecommendCount : maximum amount of items that will be recommended.

  • aAggregMethod : users' aggregation method:

    • amMode : refers to the most frequent value in a dataset.

    • amWeightedAverage : average calculated by assigning different weights to each value in the dataset, based on its importance.

    • amSimpleSum : sum of all values in the dataset, without applying any weights or additional factors.

  • aDistanceMethod :

    • dmManhattan : measures the distance between two points by summing the absolute differences in each dimension, as if traveling along straight lines on a grid.

    • dmEuclidean : calculates the straight-line distance between two points in space, being the square root of the sum of the squares of the differences between the coordinates.

    • dmCosine : measures the similarity between two vectors based on the cosine of the angle between them, focusing on the direction, not the magnitude.

    • dmJaccard : evaluates the similarity between sets by dividing the size of the intersection by the size of the union of the sets.

    • dmPearson : measures the linear correlation between two variables, evaluating the strength and direction of the relationship between them.

  • aCalculateItemDistanceOnCreate : defines if the item distance matrix will be calculated at creation time, used only for item-based recommendation. If not calculated at creation, it will be computed during the first item prediction.

  • aHasHeader : indicates whether the file has a header.

• function RecommendFromItem(aItemID: Integer): TArray; : recommends X items based on a given registered item.

• function RecommendFromItem(aItemInfo: TArray; aIDSearch : Integer = -1): TArray; : recommends X items based on the data of an item that is not in the dataset.

• function RecommendFromUser(aUserID: Integer): TArray; : recommends X items based on a given registered user.

• function RecommendFromUser(aUserInfo: TArray; aIDSearch : Integer = -1): TArray; : recommends X items based on the data of a user who is not in the dataset.