Graph Co-Attentive Session-based Recommendation

Session-based recommendation aims to generate recommendations merely based on the ongoing session, which is a challenging task. Previous methods mainly focus on modeling the sequential signals or the transition relations between items in the current session using RNNs or GNNs to identify user’s intent for recommendation. Such models generally ignore the dynamic connections between the local and global item transition patterns, although the global information is taken into consideration by exploiting the global-level pair-wise item transitions. Moreover, existing methods that mainly adopt the cross-entropy loss with softmax generally face a serious over-fitting problem, harming the recommendation accuracy. Thus, in this article, we propose a Graph Co-Attentive Recommendation Machine (GCARM) for session-based recommendation. In detail, we first design a Graph Co-Attention Network (GCAT) to consider the dynamic correlations between the local and global neighbors of each node during the information propagation. Then, the item-level dynamic connections between the output of the local and global graphs are modeled to generate the final item representations. After that, we produce the prediction scores and design a Max Cross-Entropy (MCE) loss to prevent over-fitting. Extensive experiments are conducted on three benchmark datasets, i.e., Diginetica, Gowalla, and Yoochoose. The experimental results show that GCARM can achieve the state-of-the-art performance in terms of Recall and MRR, especially on boosting the ranking of the target item.

A collaborative filtering algorithm based on item labels and Hellinger distance for sparse data

The Neighbourhood-based collaborative filtering (CF) algorithm has been widely used in recommender systems. To enhance the adaptability to the sparse data, a CF with new similarity measure and prediction method is proposed. The new similarity measure is designed based on the Hellinger distance of item labels, which overcomes the problem of depending on common-rated items (co-rated items). In the proposed prediction method, we present a new strategy to solve the problem that the neighbour users do not rate the target item, that is, the most similar item rated by the neighbour user is used to replace the target item. The proposed prediction method can significantly improve the utilisation of neighbours and obviously increase the accuracy of prediction. The experimental results on two benchmark datasets both confirm that the proposed algorithm can effectively alleviate the sparse data problem and improve the recommendation results.

Approach direction and accuracy, but not response times, show spatial-numerical association in chicks

Chicks trained to identify a target item in a sagittally-oriented series of identical items show a higher accuracy for the target on the left, rather than that on the right, at test when the series was rotated by 90°. Such bias seems to be due to a right hemispheric dominance in visuospatial tasks. Up to now, the bias was highlighted by looking at accuracy, the measure mostly used in non-human studies to detect spatial numerical association, SNA. In the present study, processing by each hemisphere was assessed by scoring three variables: accuracy, response times and direction of approach. Domestic chicks were tested under monocular vision conditions, as in the avian brain input to each eye is mostly processed by the contralateral hemisphere. Four-day-old chicks learnt to peck at the 4th element in a sagittal series of 10 identical elements. At test, when facing a series oriented fronto-parallel, birds confined their responses to the visible hemifield, with high accuracy for the 4th element. The first element in the series was also highly selected, suggesting an anchoring strategy to start the proto-counting at one end of the series. In the left monocular condition, chicks approached the series starting from the left, and in the right monocular condition, they started from the right. Both hemispheres appear to exploit the same strategy, scanning the series from the most lateral element in the clear hemifield. Remarkably, there was no effect in the response times: equal latency was scored for correct or incorrect and for left vs. right responses. Overall, these data indicate that the measures implying a direction of choice, accuracy and direction of approach, and not velocity, i.e., response times, can highlight SNA in this paradigm. We discuss the relevance of the selected measures to unveil SNA.

A comparative study to recognize fake ratings in recommendation system using classification techniques

The recommendation system is affected with attacks when the users are given liberty to rate the items based on their impression about the product or service. Some malicious user or other competitors’ try to inject fake rating to degrade the item’s graces that are mostly adored by several users. Attacks in the rating matrix are not executed just by a single profile. A group of users profile is injected into rating matrix to decrease the performance. It is highly complex to extract the fake ratings from the mixture of genuine profile as it resides the same pattern. Identifying the attacked profile and the target item of the fake rating is a challenging task in the big data environment. This paper proposes a unique method to identify the attacks in collaborating filtering method. The process of extracting fake rating is carried out in two phases. During the initial phase, doubtful user profile is identified from the rating matrix. In the following phase, the target item is analysed using push attack count to reduce the false positive rates from the doubtful user profile. The proposed model is evaluated with detection rate and false positive rates by considering the filler size and attacks size. The experiment was conducted with 6%, 8% and 10% filler sizes and with different attack sizes that ranges from 0%–100%. Various classification techniques such as decision tree, logistic regression, SVM and random forest methods are used to classify the fake ratings. From the results, it is witnessed that SVM model works better with random and bandwagon attack models at an average of 4% higher accuracy. Similarly the decision tree method performance better at an average of 3% on average attack model.

Pattern-enhanced Contrastive Policy Learning Network for Sequential Recommendation

Sequential recommendation aims to predict users’ future behaviors given their historical interactions. However, due to the randomness and diversity of a user’s behaviors, not all historical items are informative to tell his/her next choice. It is obvious that identifying relevant items and extracting meaningful sequential patterns are necessary for a better recommendation. Unfortunately, few works have focused on this sequence denoising process. In this paper, we propose a PatteRn-enhanced ContrAstive Policy Learning Network (RAP for short) for sequential recommendation, RAP formalizes the denoising problem in the form of Markov Decision Process (MDP), and sample actions for each item to determine whether it is relevant with the target item. To tackle the lack of relevance supervision, RAP fuses a series of mined sequential patterns into the policy learning process, which work as a prior knowledge to guide the denoising process. After that, RAP splits the initial item sequence into two disjoint subsequences: a positive subsequence and a negative subsequence. At this, a novel contrastive learning mechanism is introduced to guide the sequence denoising and achieve preference estimation from the positive subsequence simultaneously. Extensive experiments on four public real-world datasets demonstrate the effectiveness of our approach for sequential recommendation.

Different contribution of the monkey prefrontal and premotor dorsal cortex in decision-making supported by inferential reasoning

Several studies reported similar neural modulations between brain areas of the frontal cortex, such as the dorsolateral prefrontal (DLPFC) and the premotor dorsal (PMd) cortex, in tasks requiring encoding of the abstract rules for selecting the proper action. Here, we compared the DLPFC and PMd neuronal activity of monkeys trained in choosing the highest ranking image of pair (target item), selected from an arbitrarily rank-ordered set (A>B>C>D>E>F) in the context of a transitive inference task. Once acquired by trial-and-error, the ordinal relationship between pairs of adjacent images (i.e. A>B; B>C; C>D; D>E; E>F), monkeys were tested in inferring the ordinal relation between items of the list not paired during learning. During inferential decisions, we observed that the choice accuracy increased and the reaction time decreased as the rank difference between the compared items enhanced. This result is in line with the hypothesis that after learning, the monkeys built an abstract mental representation of the ranked items, where rank comparisons correspond to the item position comparison on this representation. In both brain areas, we observed higher neuronal activity when the target item appeared in a specific location on the screen, with respect to the opposite position and that this difference was particularly enhanced at lower degrees of difficulty. By comparing the time evolution of the activity of the two areas, we revealed that the neural encoding of target item spatial position occurred earlier in DLPFC than in PMd, while in PMd the spatial encoding duration was longer.

No negative impact of word retraining on vocabulary use or clarity of communication in Semantic Dementia

Word retraining techniques can improve picture naming of treated items in people with semantic dementia (SD). The utility of this, however, has been questioned given the propensity for under- and overgeneralisation errors in naming in SD. Few studies have investigated the occurrence of such errors. This study examined whether, following tailored word retraining: 1) misuse of words increases, 2) the type of naming errors changes, and/or 3) clarity of communication is reduced. Performance on trained and untrained word naming from nine participants with SD who completed a word retraining program were analysed. Responses from baseline and post-intervention assessments were coded for misuse (i.e., trained word produced for another target item), error type, and communication clarity. All participants showed significant improvement for trained vocabulary. There was no significant increase in misuse of words, with such errors occurring rarely. At a group level, there was an increased tendency toward omission errors for untrained items, and a reduction in semantically related responses. However, this did not impact on clarity scores with no consistent change across participants. In sum, we found no negative impacts following tailored word retraining, providing further evidence of the benefit of these programs for individuals with SD.

Dynamic Graph Learning for Session-Based Recommendation

Session-based recommendation (SBRS) aims to make recommendations for users merely based on the ongoing session. Existing GNN-based methods achieve satisfactory performance by exploiting the pair-wise item transition pattern; however, they ignore the temporal evolution of the session graphs over different time-steps. Moreover, the widely applied cross-entropy loss with softmax in SBRS faces the serious overfitting problem. To deal with the above issues, we propose dynamic graph learning for session-based recommendation (DGL-SR). Specifically, we design a dynamic graph neural network (DGNN) to simultaneously take the graph structural information and the temporal dynamics into consideration for learning the dynamic item representations. Moreover, we propose a corrective margin softmax (CMS) to prevent overfitting in the model optimization by correcting the gradient of the negative samples. Comprehensive experiments are conducted on two benchmark datasets, that is, Diginetica and Gowalla, and the experimental results show the superiority of DGL-SR over the state-of-the-art baselines in terms of Recall@20 and MRR@20, especially on hitting the target item in the recommendation list.

Zero-shot search termination reveals a dissociation between implicit and explicit metacognitive knowledge

In order to infer that a target item is missing from a display, subjects must know that they would have detected it if it was present. This form of counterfactual reasoning critically relies on metacognitive knowledge about spatial attention and visual search behaviour. Previous work on visual search established that this knowledge is constructed and expanded based on task experience. Here we show that some metacognitive knowledge is also available to participants in the first few trials of the task, and that this knowledge can be used to guide decisions about search termination even if it is not available for explicit report.