Deep Integration of Health Information Service System and Data Mining Analysis Technology

Computer software technology refers to the program or related data set up in order to ensure the normal operation of the computer. Software is the interface between users and hardware. It is the core component of maintaining the normal operation of the computer. It is also the channel of communication between users and computers. It can improve the overall structure of the computer, meticulousness and reliability [6,7,8]. Computer software Technology is a type of computer technology, including data processing, artificial intelligence, process control and scientific calculation. It has the following characteristics: (1)

The cost of development continues to increase.

Data Mining is the process of mining required data from large amounts of data stored in databases, data warehouses, or other storage information bases. It is generally accepted that data mining is a process of extracting hidden, potential, effective, novel, useful and ultimately understandable knowledge from many incomplete, noisy, fuzzy and random data [9, 10]. The general flow of data mining is shown in Figure 2.

Fig. 2

Currently, every country has been strengthening the construction of information resources services and standardising the content of information services continuously. Based on the research of other scholars, this paper builds an accurate health knowledge base, mainly from four aspects to ensure its accuracy. As shown in Figure 3, there is a multi-faceted and authoritative source of health information. We develop a set of web crawler technology to update the knowledge base in real time, select content scoring tools to measure the quality of information, and integrate heterogeneous digital health resources.

Fig. 3

Health knowledge base has a wide range of information sources to extract more accurate and valuable health information from more comprehensive information. Faced with numerous health information sources, this study designed a web crawler program to obtain the latest health information released by various information sources in time. Current web crawler technologies include: General Purpose Web Crawler, Focused Web Crawler, Incremental Web Crawler and Deep Web Crawler. ‘General Purpose Web Crawler’ crawls a wide range, a large number, and is generally used to crawl search engines, but due to commercial reasons, it is not completely open to the public; ‘Focused Web Crawler’ is a technology applied on the premise of identifying crawling topics, so it is also known as ‘Theme Network’. Because the purpose of crawling is very clear, it only crawls the web resources related to the theme, so it has fast crawling speed and less hardware occupation; ‘Incremental Web Crawler’ is to crawl new pages after determining the crawl, which is more real-time than periodic crawling; ‘Deep Web Crawler’ is a needle. For some deep web pages, such as those accessible only by using user accounts, there is a requirement for the user's right of access to the web pages. Most of the information sources of the health knowledge base constructed in this study are the websites of some government agencies and universities. So the amount of information released in a single day will not be much, the task of crawling is not very large, and the process control of crawling will not be cumbersome. Combining with the characteristics of this study, the strategy of ‘Focused Web Crawler + Incremental Web Crawler’ is selected to crawl health information sources.

As the part of data integration, this research integrates three technologies, XML, Web Service and Message Middleware, to integrate heterogeneous data. First, the heterogeneous data sources are shielded by message middleware technology. Then, the standard XML format data is generated by client software. Then Web Service is used for data integration. The processed data is output by middleware for storage and invocation.

Decision tree learning is one of the most widely used classification algorithms in the field of data mining. It is a method of approximating discrete-valued functions. It has good robustness to noise data and its rules are expressed in disjunctive expressions which are easy to understand. Since the 1960s, decision tree method has been widely used in classification, prediction, rule extraction and other fields. Especially after Quilan proposed ID3 algorithm, it has been further applied and developed in the field of machine learning and knowledge discovery. Further, the representative decision tree methods include CART, SLIQ and SPRINT. In this paper, we adopt simple ID3 algorithm. In ID3 algorithm, a significant improvement of CLS algorithm is that it determines ‘certain rules of selecting test attributes’ in CLS algorithm as selecting test attributes based on information gain. In ID3 algorithm, the window method is introduced for incremental learning to solve the problem that when the training instance set is too large, all data cannot be put into memory.

The flow chart of the algorithm is shown i n Figure 2. Let T be a set of t data samples. Its class attributes can take n different values, corresponding to n different classes C_i,i ∈ (1,2,3,...,n). Assuming that S_i is the number of samples in category C_i, the amount of information needed to classify a given sample data object is: (1)I(t1,t2,⋯,tm)=−∑i=1mpilog2(pi)I({t_1},{t_2}, \cdots ,{t_m}) = - \sum\limits_{i = 1}^m {p_i}\mathop {\log }\nolimits_2 ({p_i})

Among them, p_i is the probability that any sample belongs to C_i, and it is advantageous to S_i/S calculation.

Let attribute A have k different values (a₁,a₂,...,a_k) and attribute A divide T into k subsets (S₁,S₂,...,S_k), where samples contained in T_j have a value a_j(j = 1,2,...,k) on A, if attribute A is selected as the test attribute (i.e., the best partitioning attribute), then these subsets correspond to the branches growing from the set S. Let T_ij be the sample number of class C_i in subset T_j. It is known ‘the smaller the entropy, the higher the purity of subset partition’. Entropy of subset partitioned according to A or expected information is given by the following formula: (2)E(A)=∑j=1νt1j+…+tmjt(t1j,…,tmj)E(A) = \sum\limits_{j = 1}^\nu {{{t_{1j}} + \ldots + {t_{mj}}} \over t}({t_{1j}}, \ldots ,{t_{mj}})(3)I(t1j,…,tmj)=−∑i=1npijlog2(pij)I({t_{1j}}, \ldots ,{t_{mj}}) = - \sum\limits_{i = 1}^n {p_{ij}}\mathop {\log }\nolimits_2 ({p_{ij}})

The encoding information to be obtained by branching on A is: (4)Gain(A)=I(t1,t2,…,tm)−E(A)Gain(A) = I({t_1},t2, \ldots ,{t_m}) - E(A)

That is to say, Gain is considered to be the reduction of information entropy obtained by dividing the sample set according to the value of attribute A. ID3 algorithm calculates the information gain of each attribute and selects the attribute with the largest information gain as the test attribute of a given set S, and decides to generate the corresponding branch nodes. The generated nodes are marked as the corresponding attributes, and the corresponding branches are generated according to the different values of this attribute. Each branch represents a divided sample subset. The information gain of attributes describes the concept that for a training sample set T, the amount of information needed to identify the category of a sample from T after partitioning T with an attribute A is reduced. Through it, the ability of training sample set for attribute classification can be measured. As shown in Figure 4, the algorithm flow chart.

Fig. 4

In an engineering project with large software, the defect report in software system should be described in detail and in total. Bug in software may have the following attributes: report source, author and responsible person, submission time and solution time, report content and description information, notes, priority, severity and Bug's current status. Using Bugzilla's export function, we can get the explanations of the selected attributes of all Bug report XML files as shown in Table 1, where the path relationship of the code modification function in the table is shown in Table 2.

Decision tree learning method is generally easy to transform into if-then rules, and classification rules are easy to understand. After the above analysis, we use ID3 decision tree learning method to construct classifier, which can achieve better classification effect in theory. Although the attributes of continuous values can be dealt with in decision tree learning algorithm, the learning efficiency and classification accuracy can be improved by discretising their values. The selection of discrete intervals is shown in Table 3. The values of each attribute are mapped to 0, 1, 2.

To help software maintainers to quickly understand the running status of the current system from the software version system and defect tracking system, we classify the bug repair reports of the current software system stored in Bugzilla into two categories: fixed-Bug and potential fixed-Bug.