Semantic Feature Vectors Research Articles

Service similarity measurement integrating Bi-LSTM contextual representation and attention mechanism for web service discovery

With the rapid growth and wide adoption of web services, service discovery is becoming commonly used to locate the optimal services that can meet the requirements of users. This paper proposes a bidirectional long short-term memory (Bi-LSTM) service discovery method that integrates contextual representation and attention mechanisms for web service discovery. To conduct the study, FastText word embedding with the N-gram feature is primarily used as the input of the Bi-LSTM model to obtain the context feature vector of the service function description. Moreover, the attention mechanism is employed to calculate the similarity of the service function description to gain the semantic feature vector of the correlation degree between services. Then, the multilayer perceptron network is used to mine the mapping relationship between semantic feature vectors and matching levels among services. Finally, for a given query, target services are retrieved by ranking candidate services according to the prediction cores through the matching method. The proposed method is examined against multiple evaluation metrics, including accuracy and error, and compared with the state-of-the-art searching approaches. The experimental results show that the proposed method is more effective and attains better relevant web services, which helps the request service find more accurate results. Furthermore, this study provides new insights for researchers into web service discovery.

Fishing Vessel Type Recognition Based on Semantic Feature Vector

Identifying fishing vessel types with artificial intelligence has become a key technology in marine resource management. However, classical feature modeling lacks the ability to express time series features, and the feature extraction is insufficient. Hence, this work focuses on the identification of trawlers, gillnetters, and purse seiners based on semantic feature vectors. First, we extract trajectories from massive and complex historical Vessel Monitoring System data that contain a large amount of dirty data and then extract the semantic features of fishing vessel trajectories. Finally, we input the semantic feature vectors into the LightGBM classification model for classification of fishing vessel types. In this experiment, the F1 measure of our proposed method on the East China Sea fishing vessel dataset reached 96.25, which was 6.82% higher than that of the classical feature-modeling method based on fishing vessel trajectories. Experiments show that this method is accurate and effective for the classification of fishing vessels.

RoBERTa-GCN: A New Method for Relation Extraction in Automobile Accessory Domain

The automotive industry's rapid expansion has sparked increasing interest in the realm of automotive accessories. Navigating vast information landscapes to find accurate matches has become paramount. Leveraging cutting-edge information technologies, such as knowledge graphs and graph database-based question-answering systems, offers a crucial avenue for enhancing search efficiency. Addressing challenges posed by the domain's specialized terminology and intricate relationships, this paper introduces an innovative approach that combines a pre-trained model (RoBERTa) with graph convolutional networks (GCN). Initially, the text undergoes processing through the pre-trained model, yielding semantic feature vectors that enhance comprehension of industry-specific terminology. Subsequently, a graph convolutional network (GCN) is employed to process these semantic vectors, capturing a broader scope of neighboring vector node information. This approach not only strengthens the relationships between semantic information but also captures the intricate interconnections among entities. Ultimately, an automotive accessory query knowledge graph question-answering system is constructed using extracted entity relationship triplets. Experimental results demonstrate that the proposed RoBERTa-GCN model outperforms other baseline models, achieving an impressive F1 score of 83.93%. This research significantly enhances query capabilities and exhibits versatility in handling natural language inputs from diverse users.

A study on intelligent translation of English sentences by a semantic feature extractor

Abstract In order to enhance the performance of machine translation, this article briefly introduced algorithms that can be used to extract semantic feature vectors. Then, the aforementioned algorithms were integrated with the encoder–decoder translation algorithm, and the resulting algorithms were subsequently tested. First, the performance of the semantic recognition of the long short-term memory (LSTM)-based semantic feature extractor was tested, followed by a comparison with the translation algorithm that does not include semantic features, as well as the translation algorithm that incorporates convolutional neural network-extracted semantic features. The findings demonstrated that the LSTM-based semantic feature extractor accurately identified the semantics of the source language. The proposed translation algorithm, which is based on LSTM semantic features, achieved more accurate translations compared to the other two algorithms. Furthermore, it was less affected by the length of the source language.

Semantic feature norms: a cross-method and cross-language comparison

The ability to assign meaning to perceptual stimuli forms the basis of human behavior and the ability to use language. The meanings of things have primarily been probed using behavioral production norms and corpus-derived statistical methods. However, it is not known to what extent the collection method and the language being probed influence the resulting semantic feature vectors. In this study, we compare behavioral with corpus-based norms, across Finnish and English, using an all-to-all approach. To complete the set of norms required for this study, we present a new set of Finnish behavioral production norms, containing both abstract and concrete concepts. We found that all the norms provide largely similar information about the relationships of concrete objects and allow item-level mapping across norms sets. This validates the use of the corpus-derived norms which are easier to obtain than behavioral norms, which are labor-intensive to collect, for studies that do not depend on subtle differences in meaning between close semantic neighbors.

Sheep face image dataset and DT-YOLOv5s for sheep breed recognition

Automatic identification of breeds of sheep can be valuable to the sheep industry. Due to the lack of open source datasets of multiple breeds in some researches, which leads to poor robustness of models for automatic sheep breed recognition. This paper proposes a sheep face image dataset of 10 breeds, which consists of 5200 original images of 823 sheep from seven farms in Inner Mongolia. The dataset is expanded to 44,747 images after data enhancement. A sheep breed recognition model DT-YOLOv5s based on knowledge distillation is proposed for the dataset. The model transfers the knowledge of sheep face features learned by the teacher network YOLOv5x, which has a large number of parameters but high recognition accuracy, to the lightweight student network YOLOv5s through a 15-dimensional high-dimensional semantic feature vector, so as to improve the student network's PANet feature extraction capability and recognition accuracy, achieving the purpose of lightweight and accurate recognition of the overall network model. The experimental results show that the accuracy of our proposed model is 92.75%, which is 11.19%, 13.1% and 13.65% higher than that of SSD, Faster R-CNN and YOLOv5s models, respectively. The mean average precision (mAP@[0.5:0.95]) is 94.67%, which is 7.85%, 2.35% and 4.83% higher than that of SSD, Faster R-CNN and YOLOv5s models, respectively. Meanwhile, the inference time of the model is 12.63 ms, which is 33.9 ms, 53.13 ms and 74.61 ms less than SSD, Faster R-CNN and YOLOv5x, respectively. The proposed DT-YOLOv5s model in this paper has good robustness and generalization ability, which can provide technical support for the real-time accurate identification of the sheep breed for farmers.

Semantic Similarity-Based Mobile Application Isomorphic Graphical User Interface Identification

Applying robots to mobile application testing is an emerging approach to automated black-box testing. The key to supporting automated robot testing is the efficient modeling of GUI elements. Since the application under testing often contains a large number of similar GUIs, the GUI model obtained often contains many redundant nodes. This causes the state space explosion of GUI models which has a serious effect on the efficiency of GUI testing. Hence, how to accurately identify isomorphic GUIs and construct quasi-concise GUI models are key challenges faced today. We thus propose a semantic similarity-based approach to identifying isomorphic GUIs for mobile applications. Using this approach, the information of GUI elements is first identified by deep learning network models, then, the GUI structure model feature vector and the semantic model feature vector are extracted and finally merged to generate a GUI embedding vector with semantic information. Finally, the isomorphic GUIs are identified by cosine similarity. Then, three experiments are conducted to verify the generalizability and effectiveness of the method. The experiments demonstrate that the proposed method can accurately identify isomorphic GUIs and shows high compatibility in terms of cross-platform and cross-device applications.

MOVING OBJECT DETECTION USING SEMANTIC CONVOLUTIONAL FEATURES

Moving object detection from aerial images remains an unsolved problem in computer vision research domain. Detection results are not precise due to blurry aerial images, thin edges and noise. Various methods were previously proposed for moving object detection which could not provide robust results due many challenges, i.e., noise, motion detection, lack of appropriate features, lack of effective classification approach, complex background and variations in illumination. This research proposes an efficient method for moving object detection using convolutional semantic features from VGG-16 to use motion patterns to facilitate detection in each frame and provides smaller area as region of interest. Proposed method reduces probability motion intensity information getting lost in case of same coloured object in the background and thus minimizes background complexity. After that, proposed method performs semantic features distance measurement to calculate linear distances in each frame. In this context, if there is any frame loss due to noise or illumination variation, proposed method uses Kalman filter to process that frame by illuminating noise. Finally, decision for final moving detection is determined using random forest classifier from semantic convolutional feature vector by generating a set of probabilities for each class. Experimental results show that the proposed method can detect moving objects efficiently, which in turn will decrease the operating time and increase the detection rate compared to previous research methods.

Lymphoma segmentation from 3D PET-CT images using a deep evidential network

An automatic evidential segmentation method based on Dempster-Shafer theory and deep learning is proposed to segment lymphomas from three-dimensional Positron Emission Tomography (PET) and Computed Tomography (CT) images. The architecture is composed of a deep feature-extraction module and an evidential layer. The feature extraction module uses an encoder-decoder framework to extract semantic feature vectors from 3D inputs. The evidential layer then uses prototypes in the feature space to compute a belief function at each voxel quantifying the uncertainty about the presence or absence of a lymphoma at this location. Two evidential layers are compared, based on different ways of using distances to prototypes for computing mass functions. The whole model is trained end-to-end by minimizing the Dice loss function. The proposed combination of deep feature extraction and evidential segmentation is shown to outperform the baseline UNet model as well as three other state-of-the-art models on a dataset of 173 patients.

Semantics Convolutional Neural Network for Medical Images Analysis

Big Data Analysis is a solution that makes it possible to extract valuable information from the mass of data by using deep learning algorithms and especially the Convolutional Neural Network algorithm. In this article, we have proposed an approach that allows the addition of the semantic aspect in the classification layer of the Convolutional Neural Network algorithm. The proposed approach helps medical professionals to develop an automatic system for identifying various classes of lung cancers. First, the input data are processed to reduce the search space, and the image noise, and normalize data. Then, the preprocessed data are analyzed to reduce image space by preserving all important features. After that, the semantic memory method converts the feature vectors from the analysis layer into semantic feature vectors. Finally, the last layer classifies the input image into two classes. We evaluate our approach using the LUNA16 dataset. Our study led to better results and predictions by reducing false negatives and positives using the Semantic Convolutional Neural Network algorithm. In our approach, cancer tissues can be identified with a maximum of 97.27% for accuracy and 99.46% for AUC. This model has increased efficiency compared with state-of-the-art approaches.

Clinical Application of Early Warning Scoring Based on BiLSTM-Attention in Emergency Obstetric Preexamination and Triage.

Maternity is a special category of population and the criteria for emergency prescreening cannot be directly applied to adults. Therefore, a set of criteria for grading maternal conditions should be established. In this paper, we have combined the semantic analysis technique of BiLSTM-Attention neural network and fuzzy defect risk assessment method, to develop a hybrid approach, to preprocess the text of emergency obstetric prescreening information. Furthermore, we have used word2vec to characterize the word embedding vector and highlight the features related to the degree of defects of emergency obstetric prescreening information through the attention mechanism and obtain the semantic feature vector of the warning information. BiLSTM-Attention neural network has the dual advantages of extracting bidirectional semantic information and giving weight to important judgment information which has effectively improved the semantic understanding accuracy. Experimental tests and application analysis show that the judgment model which is based on proposed method has accurately classified and graded the defects of emergency obstetric prescreening alerts. Additionally, the accuracy and microaverage value are used as evaluation indexes.

Zero-shot image classification using coupled dictionary embedding

Zero-shot learning (ZSL) is a framework to classify images that belong to unseen visual classes using their semantic descriptions about the unseen classes. We develop a new ZSL algorithm based on coupled dictionary learning. The core idea is to enforce the visual features and the semantic attributes of an image to share the same sparse representation in an intermediate embedding space, modeled as the shared input space of two sparsifying dictionaries. In the ZSL training stage, we use images from a number of seen classes for which we have access to both the visual and the semantic attributes to train two coupled dictionaries that can represent both the visual and the semantic feature vectors of an image using a single sparse vector. In the ZSL testing stage and in the absence of labeled data, images from unseen classes are mapped into the attribute space by finding the joint-sparse representations using solely the visual dictionary via solving a LASSO problem. The image is then classified in the attribute space given semantic descriptions of unseen classes. We also provide attribute-aware and transductive formulations to tackle the “domain-shift” and the “hubness” challenges for ZSL, respectively. Experiments on four primary datasets using VGG19 and GoogleNet visual features, are provided. Our performances using VGG19 features are 91.0%, 48.4%, and 89.3% on the SUN, the CUB, and the AwA1 datasets, respectively. Our performances on the SUN, the CUB, and the AwA2 datasets are 57.0%,49.7%, and 71.7%, respectively, when GoogleNet features are used. Comparison with existing methods demonstrates that our method is effective and compares favorably against the state-of-the-art. In particular, our algorithm leads to decent performance on the all four datasets.22Early partial results of this paper is presented at 2018 AAAI (Kolouri, Rostami, Owechko, & Kim, 2018).

Emergency Semantic Feature Vector Extraction From WiFi Signals for In-Home Monitoring of Elderly

The elderly population worldwide is growing drastically. Most elders prefer to live independently in their own homes, which makes them more vulnerable to emergencies and leads to higher death rates. Timely detection and notification of emergencies can minimize the impact of the adversity. In this paper, we try to extract the Emergency Semantic Feature Vector ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ESFV$</tex-math></inline-formula> ) from the ubiquitous WiFi signals in indoor environments for non-intrusive, cost-effective monitoring of the elderly and effective emergency notification. Common emergencies of the elderly are analyzed and an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ESFV$</tex-math></inline-formula> is defined as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ESFV = ({\lbrace \pi \rbrace,\lbrace \beta \rbrace, \lbrace \rho \rbrace })$</tex-math></inline-formula> , which is a combination of the <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</u> osition, (abnormal) <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</u> ehavior and (abnormal) <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</u> espiration of the elderly. To extract the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ESFV$</tex-math></inline-formula> from WiFi signals, we design a WiFi signal structuring approach, which converts WiFi signals into two structured features, i.e., keypoint maps and respiration graphs, using WiFi Channel State Information (CSI). For keypoint map construction, we extract human-related WiFi signals, based on which we generate CSI maps. Meanwhile, a synchronized camera is used to extract human keypoints and annotate WiFi signals. A neural network is designed to transform the CSI maps into keypoint maps. For respiration graph extraction, we design a respiration-related WiFi signal extraction method and a subcarrier selection algorithm to obtain the respiration graphs. Then we extract the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \pi \rbrace$</tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \beta \rbrace$</tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \rho \rbrace$</tex-math></inline-formula> from the two structured features: <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \pi \rbrace$</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \beta \rbrace$</tex-math></inline-formula> are extracted from the keypoint maps while <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\lbrace \rho \rbrace$</tex-math></inline-formula> is extracted from the respiration graphs. Finally, the feature vector <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$(\lbrace \pi \rbrace,\lbrace \beta \rbrace, \lbrace \rho \rbrace)$</tex-math></inline-formula> of an emergency semantic is formed for notification. We build a prototype system and conduct extensive experiments. Experiment results show that our approaches perform well in both signal structuring (keypoint map construction and respiration graph extraction) and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ESFV$</tex-math></inline-formula> extraction. These demonstrate the effectiveness of our approaches for in-home monitoring and emergency notification.

Sentiment Weighted Word Embedding for Big Text Data

Sentiment analysis is the practice of eliciting a sentiment orientation of people's opinions (i.e. positive, negative and neutral) toward the specific entity. Word embedding technique like Word2vec is an effective approach to encode text data into real-valued semantic feature vectors. However, it fails to preserve sentiment information that results in performance deterioration for sentiment analysis. Additionally, big sized textual data consisting of large vocabulary and its associated feature vectors demands huge memory and computing power. To overcome these challenges, this research proposed a MapReduce based Sentiment weighted Word2Vec (MSW2V), which learns the sentiment and semantic feature vectors using sentiment dictionary and big textual data in a distributed MapReduce environment, where memory and computing power of multiple computing nodes are integrated to accomplish the huge resource demand. Experimental results demonstrate the outperforming performance of the MSW2V compared to the existing distributed and non-distributed approaches.

TransExplain: Using neural networks to find suitable explanations for Chinese phrases

When reading an article, especially a professional article, we often encounter words or phrases that we don't recognize. They may be specific domain terms or emerging entities. When we can't guess their meaning from the article, we generally refer to the terminology dictionary or search for related content on the Internet to understand them. Some researchers have used natural language generation (NLG) models to explain these unknown phrases in recent years automatically. Still, current NLG models have difficulties generating long sentences with good coherence, and they are difficult to generate multiple sentences that describe unknown phrases from different angles. Therefore, this paper proposes a model that can judge whether an existing sentence can explain a certain phrase, called TransExplain. TransExplain can use LSTM, convolutional neural network, and attention mechanism to extract multiple sentence features and map them to a fixed-dimensional semantic feature vector. By calculating the cosine similarity between the semantic feature vector and the unknown phrase vector, it is judged whether the sentence can explain the semantics of the unknown phrase. And a loss function called positive and negative means square error is introduced to improve the model's ability that distinguishes negative examples. For this task, we provided a Chinese dataset containing phrases and explanation pairs in 7 important domains. On this dataset, TransExplain can achieve better results than previous similar tasks.

Toward more accurate developer recommendation via inference of development activities from interaction with bug repair process

AbstractSoftware projects usually receive a large number of submitted bug reports every day. Manually triaging the bug reports is often time‐consuming and error‐prone; thus, it is necessary to automatically assign the bug reports to the suitable developers for bug repair, with the help of bug tracking systems. Aiming to reducing the time consumption and mismatch of bug report assignments, we present a developer recommendation model for bug repair based on weighted recurrent neural network, namely, DTPM, which contains two parts: One obtains multisource semantic information of bug reports and fuses them into high‐dimensional semantic feature vectors, and the other combines a penalty matrix into a single hidden layer neural network to obtain more reasonable developer recommendations. We conduct experiments on five datasets of open bug repositories (NetBeans, OpenOffice, GCC, Mozilla, and Eclipse), and the experimental results show that DTPM can achieve better performance than state‐of‐the‐art models LDA_KL, LDA_KL, LDA_SVM, DERTOM, DREX, and DeepTriage.

Social Relationship Inference Over Private Vehicle Mobility Data

The ubiquity of private vehicles with positioning services leaves a great deal of mobility data in the physical world, which supports abundant mobile applications in the Internet of Vehicles. Despite numerous desirable features that the data provide, the social relationship privacy inherent in private vehicle mobility data has gone with little notice. The relevant work that concentrates on social relation privacy either only considers the temporal and spatial features, attempts to obtain the explicit venue cooccurrence frequency statistics of mobility data, or characterizes the semantics of locations by directly matching the POI information to the location. In this paper, we propose a SEmantic-aware and Memory-Efficient scheme (SEME) for inferring social relationships from private vehicle mobility data. We determine the probability distribution of the visit purpose for each stopover location in vehicle trajectories by a probabilistic generative model with a latent variable of semantic feature vectors that embeds the semantic information, time context, and correlations. Labeling the trajectories with visit purposes, we derive a mobility feature vector for each driver with a feature learning model. Based on the mobility feature vectors, the similarity score that indicates the social connection strength can be pairwise-computed with an effective measurement, i.e., the cosine similarity. Additionally, for both scalability and computational efficiency, we convert the large-scale similarity calculation to an extended version of a maximum inner-product search problem and derive the closed-form solution of binary codes which is used to approximately solve the problem. To evaluate the performance of SEME, we conduct experiments with respect to effectiveness, efficiency, and robustness based on real-world private vehicle trajectories. The evaluation results demonstrate that our scheme improves the inference accuracy and reduces the time cost of similarity calculation.

Ternary Sentiment Classification of Airline Passengers’ Twitter Text Based on BERT

Traditional word vector generation model cannot solve the polysemy problem of word representation in Twitter text, so, a method is presented here, firstly, by using Bidirectional Encoder Representations from Transformers (BERT), the semantic feature vector of the text can be obtained, and then, the feature vector is inputted into the Softmax classifier to implement the sentiment classification of Twitter text. The experimental datasets are source from passengers’ Twitter comments of USA six airlines, the sentiment classification model based on Embeddings from Language Model (ELMo) as the experimental control group. The experimental results indicate that the proposed model is advantage over the experimental control group by using F1-score as the evaluation index.

Truncation Cross Entropy Loss for Remote Sensing Image Captioning

Recently, remote sensing image captioning (RSIC) has drawn an increasing attention. In this field, the encoder–decoder-based methods have become the mainstream due to their excellent performance. In the encoder–decoder framework, the convolutional neural network (CNN) is used to encode a remote sensing image into a semantic feature vector, and a sequence model such as long short-term memory (LSTM) is subsequently adopted to generate a content-related caption based on the feature vector. During the traditional training stage, the probability of the target word at each time step is forcibly optimized to 1 by the cross entropy (CE) loss. However, because of the variability and ambiguity of possible image captions, the target word could be replaced by other words like its synonyms, and therefore, such an optimization strategy would result in the overfitting of the network. In this article, we explore the overfitting phenomenon in the RSIC caused by CE loss and correspondingly propose a new truncation cross entropy (TCE) loss, aiming to alleviate the overfitting problem. In order to verify the effectiveness of the proposed approach, extensive comparison experiments are performed on three public RSIC data sets, including UCM-captions, Sydney-captions, and RSICD. The state-of-the-art result of Sydney-captions and RSICD and the competitive results of UCM-captions achieved by TCE loss demonstrate that the proposed method is beneficial to RSIC.

An effective recommendation model based on deep representation learning

Recommender system has recently attracted a lot of attention in the information service community. Currently, most recommendation models use deep neural networks to learn user preferences for items and make the final recommendations. However, these current models have not effectively captured the deep semantic features of users and items and have not fully used the auxiliary information of the items. This may result in unsatisfactory recommendations for users’ projects. In order to solve the above problem, in this paper, a novel recommendation model called RM-DRL (Recommendation Model based on Deep Representation Learning) was proposed. It mainly consists of two modules: Information Preprocessing and Feature Representation. The former generates the user’s primitive feature vectors and the items used in the latter. The latter consists of two phases: Representation Learning for Item Features (RL-IF) and Representation Learning for User Features (RL-UF). The RL-IF takes the primitive feature vectors of the item as input and uses a multi-layer Convolutional Neural Network (CNN) to learn to accurately produce the semantic feature vector of the item through multi-task learning. In RL-UF, the user primitive feature vectors and semantic feature vectors of the user preference history, and the positive and negative items were taken as input, and a novel Attention-Integrated Gated Recurrent Unit (AIGRU) neural network was proposed to learn to accurately produce user semantic feature vector. After the Feature Representation module converges, the semantic feature vectors of the users and the items can be used to calculate the users’ preferences on the items via vector dot product. Extensive experiments on five real-world datasets show that RM-DRL remarkably outperforms state-of-the-art baselines in solving the recommendation problem.