# Evaluation

## Cross validation

Recommendation.cross_validationFunction
cross_validation(
n_folds::Integer,
metric::Type{<:RankingMetric},
topk::Integer,
recommender_type::Type{<:Recommender},
data::DataAccessor,
recommender_args...
)

Conduct n_folds cross validation for a combination of recommender recommender_type and ranking metric metric. A recommender is initialized with recommender_args and runs top-k recommendation.

source
cross_validation(
n_folds::Integer,
metric::Type{<:AccuracyMetric},
recommender_type::Type{<:Recommender},
data::DataAccessor,
recommender_args...
)

Conduct n_folds cross validation for a combination of recommender recommender_type and accuracy metric metric. A recommender is initialized with recommender_args.

source
Recommendation.leave_one_outFunction
leave_one_out(
metric::Type{<:RankingMetric},
topk::Integer,
recommender_type::Type{<:Recommender},
data::DataAccessor,
recommender_args...
)

Conduct leave-one-out cross validation (LOOCV) for a combination of recommender recommender_type and accuracy metric metric. A recommender is initialized with recommender_args and runs top-k recommendation.

source
leave_one_out(
metric::Type{<:AccuracyMetric},
recommender_type::Type{<:Recommender},
data::DataAccessor,
recommender_args...
)

Conduct leave-one-out cross validation (LOOCV) for a combination of recommender recommender_type and accuracy metric metric. A recommender is initialized with recommender_args.

source

## Rating metrics

Recommendation.RMSEType
RMSE

Root Mean Squared Error.

measure(
metric::RMSE,
truth::AbstractVector,
pred::AbstractVector
)

truth and pred are expected to be the same size.

source
Recommendation.MAEType
MAE

Mean Absolute Error.

measure(
metric::MAE,
truth::AbstractVector,
pred::AbstractVector
)

truth and pred are expected to be the same size.

source

## Ranking metrics

Let a target user $u \in \mathcal{U}$, set of all items $\mathcal{I}$, ordered set of top-$N$ recommended items $I_N(u) \subset \mathcal{I}$, and set of truth items $\mathcal{I}^+_u$.

Recommendation.RecallType
Recall

Recall-at-$k$ (Recall@$k$) indicates coverage of truth samples as a result of top-$k$ (topk) recommendation. The value is computed by the following equation:

$$$\mathrm{Recall@}k = \frac{|\mathcal{I}^+_u \cap I_N(u)|}{|\mathcal{I}^+_u|}.$$$

Here, $|\mathcal{I}^+_u \cap I_N(u)|$ is the number of true positives.

measure(
metric::Recall,
truth::AbstractVector{T},
pred::AbstractVector{T},
topk::Integer
)
source
Recommendation.PrecisionType
Precision

Precision-at-$k$ (Precision@$k$) evaluates correctness of a top-$k$ (topk) recommendation list $I_N(u)$ according to the portion of true positives in the list as:

$$$\mathrm{Precision@}k = \frac{|\mathcal{I}^+_u \cap I_N(u)|}{|I_N(u)|}.$$$
measure(
metric::Precision,
truth::AbstractVector{T},
pred::AbstractVector{T},
topk::Integer
)
source
Recommendation.MAPType
MAP

While the original Precision@$N$ provides a score for a fixed-length recommendation list $I_N(u)$, mean average precision (MAP) computes an average of the scores over all recommendation sizes from 1 to $|\mathcal{I}|$. MAP is formulated with an indicator function for $i_n$, the $n$-th item of $I(u)$, as:

$$$\mathrm{MAP} = \frac{1}{|\mathcal{I}^+_u|} \sum_{n = 1}^{|\mathcal{I}|} \mathrm{Precision@}n \cdot \mathbb{1}_{\mathcal{I}^+_u}(i_n).$$$

It should be noticed that, MAP is not a simple mean of sum of Precision@$1$, Precision@$2$, $\dots$, Precision@$|\mathcal{I}|$, and higher-ranked true positives lead better MAP.

measure(
metric::MAP,
truth::AbstractVector{T},
pred::AbstractVector{T}
)
source
Recommendation.AUCType
AUC

ROC curve and area under the ROC curve (AUC) are generally used in evaluation of the classification problems, but these concepts can also be interpreted in a context of ranking problem. Basically, the AUC metric for ranking considers all possible pairs of truth and other items which are respectively denoted by $i^+ \in \mathcal{I}^+_u$ and $i^- \in \mathcal{I}^-_u$, and it expects that the $best'' recommender completely ranks$i^+$higher than$i^-, as follows:

AUC calculation keeps track the number of true positives at different rank in $\mathcal{I}$. At line 8, the function adds the number of true positives which were ranked higher than the current non-truth sample to the accumulated count of correct pairs. Ultimately, an AUC score is computed as portion of the correct ordered $(i^+, i^-)$ pairs in the all possible combinations determined by $|\mathcal{I}^+_u| \times |\mathcal{I}^-_u|$ in set notation.

measure(
metric::AUC,
truth::AbstractVector{T},
pred::AbstractVector{T}
)
source
Recommendation.ReciprocalRankType
ReciprocalRank

If we are only interested in the first true positive, reciprocal rank (RR) could be a reasonable choice to quantitatively assess the recommendation lists. For $n_{\mathrm{tp}} \in \left[ 1, |\mathcal{I}| \right]$, a position of the first true positive in $I(u)$, RR simply returns its inverse:

$$$\mathrm{RR} = \frac{1}{n_{\mathrm{tp}}}.$$$

RR can be zero if and only if $\mathcal{I}^+_u$ is empty.

measure(
metric::ReciprocalRank,
truth::AbstractVector{T},
pred::AbstractVector{T}
)
source
Recommendation.MPRType
MPR

Mean percentile rank (MPR) is a ranking metric based on $r_{i} \in [0, 100]$, the percentile-ranking of an item $i$ within the sorted list of all items for a user $u$. It can be formulated as:

$$$\mathrm{MPR} = \frac{1}{|\mathcal{I}^+_u|} \sum_{i \in \mathcal{I}^+_u} r_{i}.$$$

$r_{i} = 0\%$ is the best value that means the truth item $i$ is ranked at the highest position in a recommendation list. On the other hand, $r_{i} = 100\%$ is the worst case that the item $i$ is at the lowest rank.

MPR internally considers not only top-$N$ recommended items also all of the non-recommended items, and it accumulates the percentile ranks for all true positives unlike MRR. So, the measure is suitable to estimate users' overall satisfaction for a recommender. Intuitively, $\mathrm{MPR} > 50\%$ should be worse than random ranking from a users' point of view.

measure(
metric::MPR,
truth::AbstractVector{T},
pred::AbstractVector{T}
)
source
Recommendation.NDCGType
NDCG

Like MPR, normalized discounted cumulative gain (NDCG) computes a score for $I(u)$ which places emphasis on higher-ranked true positives. In addition to being a more well-formulated measure, the difference between NDCG and MPR is that NDCG allows us to specify an expected ranking within $\mathcal{I}^+_u$; that is, the metric can incorporate $\mathrm{rel}_n$, a relevance score which suggests how likely the $n$-th sample is to be ranked at the top of a recommendation list, and it directly corresponds to an expected ranking of the truth samples.

measure(
metric::NDCG,
truth::AbstractVector{T},
pred::AbstractVector{T},
topk::Integer
)
source

## Aggregated metrics

Return a single aggregated score for an array of multiple top-$k$ recommended items. Recommender Systems Handbook. gives an overview of such aggregated metrics. In particular, the formulation of Gini index and Shannon Entropy can be found at Eq. (20) and (21) on page 26.

Recommendation.AggregatedDiversityType
AggregatedDiversity

The number of distinct items recommended across all suers. Larger value indicates more diverse recommendation result overall.

measure(
metric::AggregatedDiversity, recommendations::AbstractVector{<:AbstractVector{<:Integer}}
)

Let $U$ and $I$ be a set of users and items, respectively, and $L_N(u)$ a list of top-$N$ recommended items for a user $u$. Here, an aggregated diversity can be calculated as:

$$$\left| \bigcup\limits_{u \in U} L_N(u) \right|$$$
source
Recommendation.ShannonEntropyType
ShannonEntropy

If we focus more on individual items and how many users are recommended a particular item, the diversity of topk recommender can be defined by Shannon Entropy:

$$$-\sum_{j = 1}^{|I|} \left( \frac{\left|\{u \mid u \in U \wedge i_j \in L_N(u) \}\right|}{N |U|} \ln \left( \frac{\left|\{u \mid u \in U \wedge i_j \in L_N(u) \}\right|}{N |U|} \right) \right)$$$

where $i_j$ denotes $j$-th item in the available item set $I$.

measure(
metric::ShannonEntropy, recommendations::AbstractVector{<:AbstractVector{<:Integer}};
topk::Integer
)
source
Recommendation.GiniIndexType
GiniIndex

Gini Index, which is normally used to measure a degree of inequality in a distribution of income, can be applied to assess diversity in the context of topk recommendation:

$$$\frac{1}{|I| - 1} \sum_{j = 1}^{|I|} \left( (2j - |I| - 1) \cdot \frac{\left|\{u \mid u \in U \wedge i_j \in L_N(u) \}\right|}{N |U|} \right)$$$
measure(
metric::GiniIndex, recommendations::AbstractVector{<:AbstractVector{<:Integer}};
topk::Integer
)

The index is 0 when all items are equally chosen in terms of the number of recommended users.

source

## Intra-list metrics

Given a list of recommended items (for a single user), intra-list metrics quantifies the quality of the recommendation list from a non-accuracy perspective. A Survey of Serendipity in Recommender Systems highlights the foundation of these metrics.

Recommendation.CoverageType
Coverage

Catalog coverage is a ratio of recommended items among catalog, which represents a set of all available items.

measure(
metric::Coverage, recommendations::Union{AbstractSet, AbstractVector};
catalog::Union{AbstractSet, AbstractVector}
)
source
Recommendation.NoveltyType
Novelty

The number of recommended items that have not been observed yet i.e., not in observed.

measure(
metric::Novelty, recommendations::Union{AbstractSet, AbstractVector};
observed::Union{AbstractSet, AbstractVector}
)
source
Recommendation.SerendipityType
Serendipity

Return a sum of relevance-unexpectedness multiplications for all recommended items.

measure(
metric::Serendipity, recommendations::Union{AbstractSet, AbstractVector};
relevance::AbstractVector, unexpectedness::AbstractVector
)
source