umutphp / laravel-model-recommendation Goto Github PK

The package creates item-based collaborative filtering but in some cases user-based filtering is also needed. With this issue, user-based filtering algorithm will be added as a new algorithm.

Discussed in #21

Some packages like multicaret/laravel-acquaintances uses one single table for all kind of relations (like, follow, vote etc). The relation type is defined by a field in the data table. This library should choose correct data in the subjected relation to generate the correct recommendation.

The library supports on same type of model in the relation. So, only recommendation between same type of model are created. The library should support different type of model to be recommended.

Prepare and document use cases with some Laravel packages like;

• overtrue/laravel-follow

When "similarity" is used, the score is not stored in the table. So, sorting is not working for recommendation list in similarity case.

Discussed in #15

After finishing #3 ;

• overtrue/laravel-like
• overtrue/laravel-favorite
• overtrue/laravel-subscribe

I have a table called user_views like shown bellow

And here is the code to config the recommendation i used

<?php

namespace App\Business\Recommendation;

use App\Models\Course;
use App\Models\Document;
use App\Models\Question;

trait HasViewsRecommendation
{


    public static $recommendations = [
        Document::class => 'suggested_documents',
        Course::class => 'suggested_courses',
        Question::class => 'suggested_questions'
    ];

    public static $recommendationCount = 5;
    private static function generateRelationAlgorithmArray(string $model, string $name): array
    {

        return
            [$name => [
                'recommendation_algorithm'         => 'db_relation',
                'recommendation_data_table'        => 'user_views',
                'recommendation_data_table_filter' => [],
                'recommendation_data_field'        => 'viewable_id',
                'recommendation_data_field_type'   => $model,
                'recommendation_group_field'       => 'user_id',
                'recommendation_count'             => self::$recommendationCount
            ]];
    }
    public static function getRecommendationAlgorithmArray(): array
    {
        $result = [];
        foreach (self::$recommendations as $model => $name) {
            $result = array_merge($result, self::generateRelationAlgorithmArray($model, $name));
        }
        return $result;
    }
}

And for sure I overided the the static function like this

class User extends Authenticatable implements  InteractWithRecommendation
{

    use HasFactory,
        HasRecommendation,
        HasViewsRecommendation,
  

    /**
     * The user recommendation config.
     * @return array
     */
    public static function getRecommendationConfig(): array
    {
        return self::getRecommendationAlgorithmArray();
    }

And i generated some recommendations using the laravel-model-recommendation table and here is some rows

Now based on my understanding .if the user logged in is of id 7 and i called the fuction php $user->getRecommendations('suggested_courses'); I should get the First 5 rows that are of type Courses .But the thing is that i always get an empty array .
To better undestand the situation I inspected the source code and I think it"s because of this query

The target_type is hardcoded to be the same as the source type.

Option to be used in User models.

Add configuration values to choose;

• Default recommendation list order
• Default recommendation count

The algorithm uses item based filtering with an unsupervised algorithm. We need to add multiple algorithm choices and a config option to choose which algorithm to use.

As the number of products gets bigger, the duration of the similarity calculation grows dramatically. Instead of using matrix calculation (n^2 operations for n products), we need to find a way to reduce the number of operations. Possible solutions;

1. Selecting a random product and use it to calculate the similarity scores of the products.
2. Creating an imaginary product with unique values.

Following the use case 6 to recommend posts for user based on votes, I came across this bug where when generating the data in the table it's adding the id of the subject_id for the source_id instead of using user_id from the interactions table.
Following the bug i found that at this line below, the group_field is not passed down in the foreach loop I couldn't go further but I think from there and on this field is gone.

that's the config I use, same as use case 6 except changed user class to Post::class

public static function getRecommendationConfig(): array
    {
        return [
            'recommended_posts' => [
                'recommendation_algorithm'         => 'db_relation',
                'recommendation_data_table'        => 'interactions',
                'recommendation_data_table_filter' => [
                    'relation' => 'upvote' // possible values are follow/like/subscribe/favorite/upvote/downvote. Choose the one that you want to generate the recommendation for.
                ],
                'recommendation_data_field'        => 'subject_id',
                'recommendation_data_field_type'   => Post::class,
                'recommendation_group_field'       => 'user_id',
                'recommendation_count'             => 5
            ]
        ];
    }