A recommendation system with reinforcement for determining the price of a product

Simple item page
dc.contributor.authorDrin, Svitlanaen_US
dc.contributor.authorKriuchkova, Anastasia
dc.date.accessioned2024-11-15T09:20:56Z
dc.date.available2024-11-15T09:20:56Z
dc.date.issued2023
dc.description.abstractRecommender systems belong to a fairly new concept with a wide scope of application and the property of adaptation to non-classical optimization problems. Thus, recommender systems can be used in e-commerce to predict demand and price for a new product in a chain of stores. LightGBM is a reinforcement recommendation system based on regression trees. The Boosting technology for the new tree ˆ f(x) ← ˆ f(x) + λfk(x) is selected with the specified parameter value λ. A typical value of λ is 0.001 to 0.1. But despite its simplicity and many obvious advantages, for many real big data with long-term dependence, classic Boosting techniques should not be used in real models. For example, if there are 900 stores and 7000 SKUs in the chain, then we have more than 6 million combinations per day in the event horizon. In the LightGBM technique, trees of different structure q are combined, where q : Rm → T. And in the known tree structure q(x), the optimal weight of each leaf is obtained by minimizing the loss function L′(t) = PT i=1 [ Gjwj + 1 2 (Hj + λ)w2 j ] + γT. It is worth noting that the LightGBM algorithm, which is characterized by its efficiency, accuracy and speed, creates histograms and uses the generated classes instead of the entire range of values of each variable, achieving a significant reduction in training time. The generated classes are the result of the Gradient One Side Sampling (GOSS) method. The main idea of the GOSS methodology focuses on the fact that not all observations contribute equally to the learning of the algorithm, since those with a small first derivative of the loss function learn better than those with a large first derivative of the loss function. In the case of forecasting the price of goods that do not have a sales history, we will use embedding in NLP in our model. The main purpose of using embedding is the distance between vectors of products to a new product that have a similar nature of sales, and this distance should be minimal. Results will be evaluated using the MAE performance metric.en_US
dc.identifier.citationDrin S. A recommendation system with reinforcement for determining the price of a product / Svitlana Drin, Anastasia Kriuchkova // NORDSTAT 2023, Gothenburg (19-22 June 2023). - Gothenburg, Sweden, 2023. - 1 р.en_US
dc.identifier.urihttps://ekmair.ukma.edu.ua/handle/123456789/32369
dc.language.isoenen_US
dc.relation.sourceNORDSTAT 2023, Gothenburg (19-22 June 2023)en_US
dc.statusfirst publisheden_US
dc.subjectrecommender systemsen_US
dc.subjectLightGBMen_US
dc.subjectregression treesen_US
dc.subjectMAE performance metricen_US
dc.subjectconference materialsen_US
dc.titleA recommendation system with reinforcement for determining the price of a producten_US
dc.typeConference materialsen_US
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
Drin_Kriuchkova_A_recommendation_system_with_reinforcement_for_determining_the_price_of_a_product.pdf
Size:
98.39 KB
Format:
Adobe Portable Document Format
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Кафедра математики