What is Field Embedded Factorization Machine (FEFM)?
Field Embedded Factorization Machine, or FEFM, is a type of machine learning algorithm that falls under the Factorization Machine (FM) family of algorithms. FM is used in recommendation systems, where it predicts what a user is going to like based on their past preferences. FEFM is a variant of FM that introduces symmetric matrix embeddings for each field pair along with feature vector embeddings present in FM.
How does FEFM work?
In FM, the vector embedding of the i-th feature, vi, is used. However, in FEFM, a learnable symmetric matrix WF(i), F(j) is used instead. This matrix acts as the embedding for the field pair F(i) and F(j). The interaction between the i-th and j-th feature is mediated through the learnable symmetric matrix WF(i), F(j).
FEFM uses equation phi(\theta, x) = phi_FEFM((w, v, W), x) = w0 + ∑i=1mwixi + ∑i=1m∑j=i+1mviTWF(i), F(j)vjxi xj. This equation uses a k-dimensional feature vector embedding space, where k is the dimension of the feature vector embedding space that contains feature vectors vi and vj. The symmetric property of the learnable matrix WF(i), F(j) is ensured by reparameterizing WF(i), F(j) as UF(i), F(j) + UF(i), F(j)T, where UF(i), F(j)T is the transpose of the learnable matrix UF(i), F(j). Note that W_F(i), F(j) can also be interpreted as a vector transformation matrix that transforms a feature embedding when interacting with a specific field.
What is the difference between FEFM and Field-Aware Factorization Machines (FFMs)?
FFM is another type of machine learning algorithm that explicitly learns field-specific feature embeddings, which FEFM does not. In FEFM, the learnable symmetric matrix acts as the embedding for the field pair, instead of explicitly learning field-specific feature embeddings like in FFMs. This makes FEFM more efficient and effective than FFMs in certain applications.
Furthermore, FEFM's learnable symmetric matrix can model latent correlations between features in different fields, which is not possible with FFMs. This means that FEFM is better able to capture complex relationships between different features, making it more accurate in predicting user preferences.
What are some applications of FEFM?
FEFM is most commonly used in recommendation systems, where it predicts what a user is going to like based on their past preferences. This is because FEFM is able to capture latent correlations between features in different fields, making it more accurate in predicting user preferences. It can also be used in other applications where modeling complex relationships between different features is important.
Overall, FEFM is a powerful and efficient machine learning algorithm that is able to capture complex relationships between different features in a more accurate way than other algorithms such as FFMs. This makes it especially useful in recommendation systems and other applications where predicting user preferences is important.
