Introducing a knowledge graph into a recommender system as auxiliary information can effectively solve the sparse and cold start problems existing in traditional recommender systems. In recent years, many researchers have performed related work. A recommender system with knowledge graph embedding learning characteristics can be combined with a recommender system of the following three forms: one-by-one learning, joint learning, and alternating learning. For current knowledge graph embedding, a deep learning framework only has one embedding mode, which fails to excavate the potential information from the knowledge graph thoroughly. To solve this problem, this paper proposes the Ripp-MKR model, a multitask feature learning approach for knowledge graph enhanced recommendations with RippleNet, which combines joint learning and alternating learning of knowledge graphs and recommender systems. Ripp-MKR is a deep end-to-end framework that utilizes a knowledge graph embedding task to assist recommendation tasks. Similar to the MKR model, in the Ripp-MKR model, two tasks are associated with cross and compress units, which automatically share latent features and learn the high-order interactions among items in recommender systems and entities in the knowledge graph. Additionally, the model borrows ideas from RippleNet and combines the knowledge graph with the historical interaction record of a user’s historically clicked items to represent the user’s characteristics. Through extensive experiments on real-world datasets, we demonstrate that Ripp-MKR achieves substantial gains over state-of-the-art baselines in movie, book, and music recommendations.

At present, most link prediction algorithms are based on the similarity between two entities. Social network topology information is one of the main sources to design the similarity function between entities. But the existing link prediction algorithms do not apply the network topology information sufficiently. For lack of traditional link prediction algorithms, we propose two improved algorithms: CNGF algorithm based on local information and KatzGF algorithm based on global information network. For the defect of the stationary of social network, we also provide the link prediction algorithm based on nodes multiple attributes information. Finally, we verified these algorithms on DBLP data set, and the experimental results show that the performance of the improved algorithm is superior to that of the traditional link prediction algorithm.

Minimally invasive surgery like laparoscopic surgery is an active research area of clinical practice for less pain and a faster recovery rate. Detection of surgical tools with more accurate spatial locations in surgical videos not only helps to ensure patient safety by reducing the incidence of complications but also makes a difference to assess the surgeon performance. In this paper, we propose a novel Modulated Anchoring Network for detection of laparoscopic surgery tools based on Faster R-CNN, which inherits the merits of two-stage approaches while also maintains high efficiency of comparable speed as state-ofthe-art one-stage methods. Since objects like surgical instruments with a wide aspect ratio are difficult to recognize, we develop a novel training scheme named as modulated anchoring to explicitly predict arbitrary anchor shapes of objects of interest. For taking the relationship of different tools into consideration, it is useful to embed the relation module in our network. We evaluate our method using an existing dataset (m2cai16-tool-locations) and a new private dataset (AJU-Set), both collected from cholecystectomy surgical videos in hospital, covering information of seven surgical tools with spatial bounds. We show that our detector yields excellent detection accuracy of 69.6% and 76.5% over the introduced datasets superior to other recently used architectures. We further verify the efficiency of our method by analyzing the usage patterns of tools, the economy of the movement, and the dexterity of operations to assess surgical quality. INDEX TERMS Laparoscopic surgery, tool detection, convolutional neural network, operational quality assessment.

Since the word2vec model was proposed, many researchers have vectorized the data in the research field based on it. In the field of social network, the Node2Vec model improved on the basis of word2vec can vectorize nodes and edges in social networks, so as to carry out relevant research on social networks, such as link prediction, and community division. However, social network is a network with homogeneous structure. When dealing with heterogeneous networks such as knowledge graph, Node2Vec will lead to inaccurate prediction and unreasonable vector quantization data. Specifically, in the Node2Vec model, the walk strategy for homogeneous networks is not suitable for heterogeneous networks, because the latter has distinguishing features for nodes and edges. In this paper, a Heterogeneous Network vector representation method is proposed based on random walks and Node2Vec, called KG2vec (Heterogeneous Network to Vector) that solves problems related to the inadequate consideration of the full-text semantics and the contextual relations that are encountered by the traditional vector representation of the knowledge graph. First, the knowledge graph is reconstructed and a new random walk strategy is applied. Then, two training models and optimizing strategies are proposed, so that the contextual environment between entities and relations is obtained, semantically providing a full vector representation of the Heterogeneous Network. The experimental results show that the KG2VEC model solves the problem of insufficient context consideration and unsatisfactory results of one-to-many relationship in the vectorization process of the traditional knowledge graph. Our experiments show that KG2vec achieves better performance with higher accuracy than traditional methods.