Getting similar items

Hi, not really an issue. I am new to LightFM. Before I used implicit for generating recommendations but there was no build-in option for passing in user- / item-features.

Implicit defines a function on the model for getting similar items. Does adapting this function to LightFM work? I think I’m missing the theoretical background to see for myself.

def similar_items(self, itemid, N=10):
    scores = self.item_factors.dot(self.item_factors[itemid]) / self.item_norms
    best = np.argpartition(scores, -N)[-N:]
    return sorted(zip(best, scores[best] / self.item_norms[itemid]), key=lambda x: -x[1])

https://github.com/benfred/implicit/blob/master/implicit/als.py

[EthanRosenthal]

I believe the following should work with LightFM. One thing to note is that the similarity function below ignores the item biases that LightFM learns, but I think it should be suitable for generating sensible recommendations.

def similar_items(item_id, item_features, model, N=10):
    item_representations = item_features.dot(model.item_embeddings)

    # Cosine similarity
    scores = item_representations.dot(item_representations[item_id, :])
    item_norms = np.linalg.norm(item_representations, axis=1)
    scores /= item_norms

    best = np.argpartition(scores, -N)[-N:]
    return sorted(zip(best, scores[best] / item_norms[item_id]), 
                  key=lambda x: -x[1])

[maciejkula]

This is correct. You can also use the get_<user/item>_representations methods to get the latent representations: http://lyst.github.io/lightfm/docs/lightfm.html#lightfm.LightFM.get_item_representations.