International Journal of Information Technology and Management (10 papers in press)

Regular Issues

• A Comprehensive Evaluation Method of Chinese Online Teaching Effect Based on Cluster Analysis  
  by Yuanyuan Zhang 
  Abstract: In order to improve the accuracy of evaluation results, a comprehensive evaluation method for the effectiveness of Chinese online teaching based on cluster analysis is proposed. Firstly, the AHP is used to select evaluation indicators, and the weighted average method is applied to quantify the evaluation indicators. Secondly, the K-means algorithm is used to comprehensively evaluate the teaching effectiveness, calculate the average value of all sample points within each cluster to locate the cluster centres, and iteratively update the centre position. Finally, repeat the calculation of the distance from each point to the nearest cluster centre, select new cluster centres, ensure that the distance between the initial cluster centres is as far as possible, and use the optimised K-means algorithm to achieve teaching effectiveness evaluation. The experimental results show that the proposed method has a low mean square error, indicating that its evaluation results are relatively accurate.
  Keywords: Cluster analysis; Online teaching; Effect evaluation; K-means algorithm.
  DOI: 10.1504/IJITM.2025.10070102
   
  
• The Impact of Social Media Marketing on the Purchase Behaviour of E-Commerce Consumers  
  by Yinli Hou 
  Abstract: To optimise e-commerce marketing strategies, enhance marketing effectiveness, and conduct research on the impact of social media marketing on e-commerce consumer purchasing behaviour. The article first analyses social media marketing and consumer purchasing behaviour, and then locates relevant variables based on the analysis results to establish a conceptual model of e-commerce consumer purchasing behaviour. Finally, complete the questionnaire design and construct a logistic regression model to analyse consumer online purchasing behaviour. Research has found that social media, marketing interaction objects, interactive content, and trust can have a positive impact on consumers shopping behaviour. Enterprises can optimise their social media marketing strategies, strengthen the accuracy of social positioning, enhance platform security, strengthen the authority and affinity of marketing interaction objects, enrich and enhance the detail and trust of marketing information, and thus achieve the maximisation of marketing effectiveness.
  Keywords: Social media; E-commerce; Consumers; Purchase behavior; Impact analysis.
  DOI: 10.1504/IJITM.2025.10070103
   
  
• Evaluation Method of English Course Students' Online Learning Effectiveness based on Data Mining  
  by Shuyu Li 
  Abstract: To solve the problems of poor setting of evaluation indicators and long evaluation time in existing methods, an evaluation method for online learning effectiveness of English course students based on data mining is designed in this study. Firstly, construct a dataset for online learning of English course students, and then establish an evaluation index system for the effectiveness of online learning of English course students, assigning weights to the indicators. Finally, the weighted sum of the evaluation index weights is used to obtain the online learning effectiveness score for students. The experimental results showed that after applying this method, the intra group difference of indicator data did not exceed 3.15, and the inter group difference of indicator data remained around 5.5. The maximum evaluation time during the experiment was only 12.1s, effectively achieving the design expectations.
  Keywords: Online learning data; Data feature mining; K-means algorithm; Evaluation of learning effectiveness; Indicator weight.
  DOI: 10.1504/IJITM.2025.10070105
   
  
• Secure Aggregation of Power Information Resources based on Improved Transfer Learning  
  by Ruijin He, Jinhua Huang, Huiying Lu, Xuxian Wang, Yuhao Liu, Minghui Li 
  Abstract: In order to solve the problems of long time, low accuracy, and low data integrity index of traditional methods for secure aggregation of power information resources, a secure aggregation method of power information resources based on improved transfer learning is proposed. Firstly, select transfer learning models based on the characteristics of power information resources. Secondly, identify key technologies for improving the transfer learning process and strategies for enhancing model adaptability and robustness. Finally, a secure aggregation framework for power information resources is constructed, which utilizes data generation, aggregation, and secure sharing to achieve secure aggregation of power information resources. The experimental results show that the resource aggregation time of the proposed method varies within 0.1s~0.8s, and the accuracy of power information resource aggregation can reach 90%, with a high data integrity index. This proves that the method has good security aggregation effect on power information resources.
  Keywords: Improved transfer learning; Power information resources; Secure aggregation; Data generation; Security sharing.
  DOI: 10.1504/IJITM.2025.10070106
   
  
• Fuzzy Retrieval of Power Dispatching Knowledge Base through Large Language Model Integrated with Knowledge Graph  
  by Shuhong Wu 
  Abstract: In order to solve the problems of low normalized loss accumulation gain and low retrieval consistency in traditional methods, a fuzzy retrieval method of power dispatching knowledge base through large language model integrated with knowledge graph is proposed. The entity and attribute information are carefully classified by triple algorithm and the text knowledge graph of power dispatching is constructed. StanfordNLP is used to analyse the part-of-speech of text data related to power dispatching, and the invalid triples are removed. The large language model and knowledge graph are fused to construct the power dispatching knowledge base, and the natural language is mapped by fuzzy reasoning mechanism, and the data information with the highest fitting degree is taken as the output result of retrieval. Experiments show that the normalised cumulative loss gain of this method is close to 1, the retrieval consistency is always higher than 90%, the retrieval results are reliable.
  Keywords: Large language model; Knowledge graph; Power dispatch; Knowledge base; Triple.
  DOI: 10.1504/IJITM.2025.10070107
   
  
• An Unstructured Data Retrieval Method for English Online Course Resources based on Hash Learning Algorithm  
  by Man He, Jin Xie 
  Abstract: The retrieval of unstructured data often suffers from low retrieval accuracy and inconsistent retrieval results. Therefore, this paper proposes an unstructured data retrieval method for English online course resources based on hash learning algorithm. Constructing an unstructured dataset of English online course resources and conducting preprocessing; Construct a supervised identification matrix for unstructured data and perform dimensionality reduction on the unstructured data; Using hash functions and tanh activation functions to obtain unstructured data feature binary hash codes; Establish a hash learning algorithm data retrieval model, use the distance loss function of samples to calculate the similarity between hash codes, use keywords as tag information, and achieve the retrieval of unstructured data of English online course resources. Through experimental verification, the proposed method consistently achieves a retrieval consistency of over 90%, a retrieval recall rate of over 85%, high retrieval accuracy, and reliable retrieval results.
  Keywords: Unstructured data; Data retrieval; Hash learning; English online course resources; Supervised discrimination projection.
  DOI: 10.1504/IJITM.2025.10070108
   
  
• Accurate Recommendation Method for Enterprise Product Network Marketing Information under the Background of Big Data  
  by Chang Liu 
  Abstract: Aiming to achieve personalised and precise recommendation of marketing information, a method for accurate recommendation of enterprise product network marketing information under the background of big data is proposed. Firstly, collect user information data and pre-process the data to construct a user profile that comprehensively describes user interests and preferences based on the data processing results. Secondly, a collaborative filtering algorithm based on users and items is adopted for predicting user preferences. Finally, the three-dimensional features of marketing information are obtained through the serial parallel convolutional gate valve recurrent neural network in deep learning, and combined with user profiles and preference prediction results, the matching between users and marketing information is achieved, thereby realising personalised recommendation of marketing information. The experimental results show that the proposed method has high recommendation accuracy, high user satisfaction, and high data processing efficiency, indicating its good application effect.
  Keywords: Big data; Online marketing; Information recommendation; User profile; Collaborative filtering.
  DOI: 10.1504/IJITM.2025.10070109
   
  
• Study on Personalised Adaptive Learning Teaching Course Recommendation Method based on GP-DINA  
  by Jie Luo 
  Abstract: In order to improve the accuracy and recall of learning and teaching course recommendations, this paper proposes a personalised adaptive learning and teaching course recommendation method based on GP-DINA. Firstly, categorize the learning styles of learners based on their learning behaviour information and personalise the clustering of learning styles; Then, a GP-DINA model is constructed to analyse the similarity parameters of learners, and based on this model, the similarity of learning behaviour and preference parameters of course content are estimated; Finally, determine the weights of user behaviour preferences and course prerequisite preferences; Obtain the total recommended weight through the information entropy weight allocation method, and generate a personalised adaptive learning teaching course recommendation list. The results show that the recommendation accuracy of this method is 98.1%; The recall rate can reach 99.6%, and the cumulative revenue can reach 68%, indicating that the ranking recommendation index is highly effective.
  Keywords: GP-DINA model; XMMC information extraction model; Preference weight; Information entropy; Weight allocation.
  DOI: 10.1504/IJITM.2025.10070113
   
  
• Financial Data Security Risk Perception of Cloud Platform Based on Graph Neural Network  
  by Kejing Lu 
  Abstract: To address the challenges of suboptimal precision and extended duration associated with conventional approaches to security risk detection, a financial data security risk perception of cloud platform based on graph neural network is proposed. Collect financial data from cloud platforms and construct financial data security risk perception indicators. By filling in missing values, removing outliers, and normalising data, the data is processed. The efficiency indicators are ranked in ascending order, and the cost indicators are ranked in descending order. The rank sum ratio of each indicator is calculated, and the entropy weight method is used to calculate the indicator weights. The trained graph neural network is used to achieve the perception of financial data security risks on the cloud platform. The findings from our experiments indicate that our approach achieves a security risk perception accuracy of 94.3%,with a time consumption of only 2.3s, and the application effect is good.
  Keywords: Graph neural network; Cloud platform; Financial data; security risk perception; Rank sum ratio; Entropy weight method.
  DOI: 10.1504/IJITM.2025.10070114
   
  
• A balanced scheduling method for multi-threaded tasks based on two-level parallelism between clusters and big data clustering  
  by Xian Yang, Jue Huang, Yun Zhao, Hui Tong, Shibing Chen, Yuxin Lu, Wei Cao 
  Abstract: To improve the efficiency of task scheduling and enhance the negative load-balancing effect of tasks, this paper proposes a multi-threaded task-balancing scheduling method based on two-level parallelism between clusters and big data clustering. Firstly, use fuzzy C-means clustering to group task data into multiple clusters based on feature similarity. Then, build a multi-threaded task model that allows tasks to be executed in parallel on multiple threads, achieving two-level parallel processing between and within clusters. Finally, by determining task priority and hierarchical sorting, a task scheduling manager is designed to achieve balanced task scheduling. The experiment shows that the maximum standard deviation of the load in this method is 0.05, and the maximum over-time task ratio is 0.015, indicating that this method has strong load-balancing ability and can achieve real-time processing of tasks.
  Keywords: multi-threaded model; task scheduling; balanced scheduling; fuzzy C-means clustering; task scheduling manager.
  DOI: 10.1504/IJITM.2025.10070570