AI-ordlista
Den kompletta ordlistan över AI
Sequential Recommendation
An approach to recommender systems that models the chronological order of user interactions to predict future interests. Unlike traditional methods, it captures temporal dependencies and the evolution of preferences.
Session-based Recommendation
A subfield of sequential recommendation focusing on interactions within a single user session. It doesn't require long-term history and relies on immediate behavior to generate relevant recommendations.
Markov Decision Process
A formal mathematical model for sequential decision-making where future states depend only on the current state and action taken. Applied to recommender systems to model transitions between items.
GRU4Rec
A recurrent neural network architecture based on Gated Recurrent Units (GRU) specifically designed for session-based recommendation. It uses attention mechanisms and sampling techniques to improve performance on sequential data.
BERT4Rec
A sequential recommendation model adapted from the BERT architecture that uses bidirectional transformers to capture contextual dependencies. It applies item masking as a pre-training task to learn robust sequential representations.
Transformer-based Recommendation
A class of models using the Transformer architecture to capture long-distance dependencies in interaction sequences. These models outperform traditional RNNs thanks to multi-head attention mechanisms.
Sequential Pattern Mining
The process of extracting frequent and ordered patterns in user transaction sequences. These patterns serve as the basis for understanding typical behaviors and generating recommendations based on observed trends.
Next Item Prediction
A fundamental task in sequential recommendation that involves predicting the next item a user is likely to interact with. It serves as the training objective for most modern sequential models.
Temporal Dynamics
Phenomenon where user preferences evolve over time, requiring models that capture these gradual changes. Sequential recommendation systems incorporate these dynamics to maintain the relevance of suggestions.
User Session Modeling
Technique of representing user interactions within a session as a meaningful ordered sequence. It allows capturing short-term intent and navigation context for instant recommendations.
Sequential Embedding
Vector representation of items or users that preserves sequential and contextual information of interactions. These embeddings capture both intrinsic characteristics and temporal relationships between items.
SASRec
Self-Attentive Sequential Recommendation, a pioneering model using only attention mechanisms for sequential recommendation. It effectively captures long-distance dependencies without the limitations of traditional RNNs.
NARM
Neural Attentive Recommendation Machine, a model combining RNN and attention mechanism for session-based recommendation. It captures both overall sequential behavior and main intentions of the current session.
STAMP
Short-Term Attention/Memory Priority Model, architecture focusing on recent interactions for session-based recommendation. It uses attention to weight the importance of different items in short-term memory.
Time-aware Recommendation
Approach explicitly integrating temporal information (timestamps, durations, cycles) into recommendation models. It improves accuracy by capturing temporal patterns and recency of interactions.
Sequential Cold Start
Specific challenge for sequential systems when recommending for new users or items with little or no history. Solutions include transfer learning and session-based models to mitigate this problem.
GNN-based Sequential Recommendation
Approach using Graph Neural Networks to model complex relationships between items in sequences. It effectively captures state transitions and implicit graph structures in sequential data.
Hierarchical Sequential Modeling
Technique modeling sequences at multiple levels of abstraction (sessions, days, weeks) to capture patterns at different temporal scales. It allows for a more nuanced understanding of user behavior.
Contrastive Sequential Learning
Learning paradigm using contrastive learning to improve sequential representations by maximizing the similarity between positive views of the same sequence. It enhances the model's robustness against noise and sparse data.