Evaluation of students’ IELTS writing ability based on machine learning and neural network algorithm

Abstract

With the increase of social development and international communication, more and more Chinese candidates take IELTS test as the basis of study and work. IELTS is more difficult than national CET-4 and CET-6, listening and reading are the process of students “entering” language, they accept it passively; words, writing and translation are students”’ exit “language process, which lays the foundation for” writing “. Writing is a second language and one of the main indicators of students’ ability to use English. It is not only an important way to communicate ideas, emotional expressions and cultural exchanges, but also reflects the language ability, communication ability and thinking level of English. Therefore, English writing ability is an important index of students’ English professional ability and an important aspect of students’ English professional training. This study identified the following research tasks: according to the established measurement and evaluation indicators, the students’ writing ability was evaluated and the students’ writing ability was evaluated.

Keywords

Machine learning neural network algorithm IELTS writing evaluation index

1 Introduction

With the increase of social development and international communication, more and more Chinese candidates take IELTS test as the basis of study and work. Because of different cultures and language expressions, students always encounter difficulties in language learning, and they are too focused on terms and expressions. The content of IELTS writing test is different from the requirements of CET-4 and CET-6. The text of IELTS consists of two parts: the first part is graphic text with 150 words. The information in the flow chart is described and analyzed in graphic text, and the information content is summarized and explained in a coherent and detailed way, with emphasis on the students’ ability to describe graphic information. The second part requires students to write documents according to the subject and background, and to present their views and opinions. It tests whether students have the ability to argue. not only to study the vocabulary, grammatical basis and the structure of the article, but also to test the students’ ability to obtain information and think. the writing difficulty rate of IELTS is higher than that of the national level 4 and 6 examinations.

According to the theory of Krashen and Swain, listening and reading are students’ language “input” process, they passively accept, writing is one of the main indicators of students’ ability to learn English, writing is not only an important way to communicate ideas, emotional expressions and cultural exchanges, but also reflects language ability, communication ability and ideological level. Therefore, English writing ability is an important index of students’ English professional ability and an important aspect of students’ English professional training. This study identified the following research tasks: according to the established measurement and evaluation indicators, the students’ writing ability was evaluated and the students’ writing ability was evaluated.

2 Related work

Along with the development of education in China, the literature [1] thinks that the accuracy and richness of IELTS language and vocabulary is very important to improve the quality of English. Special emphasis was placed on the need to learn how to combine lengthy and short sentences when drafting and replacing synonyms. Literature [2] focuses on the content of the ielts oral writing test. each test details the terms and practices of the relevant subjects. in the process, students gradually acquire knowledge of these subjects, thus expanding their knowledge reserves. According to the literature [3], it is necessary to take a task way in teaching IELTS so that students can acquire a language through language practice, which improves students’ language communication ability. Document [4] enable students to apply communication in IELTS teaching to reduce the influence of their mother tongue and strengthen the logic of language, that is, gradually eliminate students’ fear of English logic in IELTS teaching, and open up students’ written ideas. Enhance their understanding of science. The literature [6] based on the theory of Hanroed and Hasan on connection classification, based on the analysis of IELTS macro structure, analyzes the high standard paradigm of IELTS writing and the low score model of IELTS writing, and concludes that appropriate correlation can improve the consistency of the article. Literature [7] holds that vocabulary plays a key role in IELTS writing. Therefore, teachers should teach IELTS high frequency phrases more, with emphasis on explaining IELTS words and high frequency phrases. Documentation [8] includes many topics related to IELTS works, such as environmental protection, language diversity, family issues and school violence issues, as well as reading background documents from the Cambridge IELTS series, which are very useful to help students write compositions. Students should make them as advanced reading materials and take relevant notes to improve their language expression.

In an evaluation study of IELTS students’ writing ability, through quantitative and qualitative research on English learners, some differences in the implementation of learning strategies between winners and losers were found. Those who succeed adopt more thinking strategies than those who fail, and they pay more attention to the meaning and language form than to the art. Literature [10] focuses on the study of the writing process, including practical cognitive models, writing conceptualization strategies, individual differences between students, and changes in the writing process at different stages. Literature [11] pointed out that writing is an important part of language skills, and the evaluation of students’ writing ability requires scientific standards. By subdividing the writing evaluation criteria, the systematic, true, comprehensive and operable evaluation criteria are determined, which is helpful to improve IELTS students’ writing ability and promote college English writing education. Literature [12] points out that writing evaluation is an important part of writing education and directly affects students’ enthusiasm and attitude. traditional writing assessment methods do not help improve students’ writing skills. in order to explore multiple writing assessment models in both languages, the design and teaching practice of writing assessment models are studied, including diagnostic assessment, self-assessment, peer assessment, group assessment, teacher assessment [13]. Literature points out that feedback is an important part of English writing evaluation. The validity and accuracy of the information received directly affect the capacity building. By analyzing the responses received in the English writing evaluation, we can see the students’ English writing ability in order to improve the effectiveness of the English writing evaluation. Document [14] pointed out that writing evaluation is an important part of English writing process. In view of the rapid development of new technology, the use of big data technology to rebuild writing ability has become a major task in English teaching reform [15, 16]. This new technology provides opportunities and challenges for English writing evaluation methods.

3 Analysis of composition of students’ IELTS writing ability

English writing ability is an important part of language skills [17]. Reading and writing ability is the ability to write articles. According to different theories, the elements and definitions of English writing ability are defined. As shown in Table 1, they define the elements of English writing ability, see Table 1 for details.

Table 1
Elements and connotations of sally burgess and katie head writing test ability

Writing test ability Connotation

Mandate outcomes Use a language consistent with the communication goal and suitable for the relationship between the reader and the author.

Highlights and Creative Outputs Writing requires the author’s ability to integrate and input specific task content, such as adding notes, charts, survey data, etc. performing tasks requires not only authors to develop and organize their own language to express their views, but also to present their own views.

Grammatical structure and lexical scope Must be able to demonstrate their vocabulary and ability to use grammatical structures.

Organization and coherence The structure of articles, paragraphs and chapters must be generally understood.

Style and register It is necessary to understand the expression of different languages and the scope of the relationship between readers and authors.

Have the desired communicative effect on readers This is mainly the result of the assessment.

Writing test ability	Connotation
Mandate outcomes	Use a language consistent with the communication goal and suitable for the relationship between the reader and the author.
Highlights and Creative Outputs	Writing requires the author’s ability to integrate and input specific task content, such as adding notes, charts, survey data, etc. performing tasks requires not only authors to develop and organize their own language to express their views, but also to present their own views.
Grammatical structure and lexical scope	Must be able to demonstrate their vocabulary and ability to use grammatical structures.
Organization and coherence	The structure of articles, paragraphs and chapters must be generally understood.
Style and register	It is necessary to understand the expression of different languages and the scope of the relationship between readers and authors.
Have the desired communicative effect on readers	This is mainly the result of the assessment.

English writing ability is closely related to skills, and English writing can be divided into thought expression, language use, writing and standards [18 –20], as shown in the picture, see Fig. 1 for details.

Fig.1

Relationship between students’ academic knowledge and IELTS writing ability.

After the merger, the words “content” and “structure” correspond to the level of thought and expression; the use of grammar at the level of language application should be considered to meet United Nations standards [21, 22]. Therefore, the conclusion of this study is to evaluate three aspects of English “writing” ability: English thought and expression, language use and language standard writing. Therefore, these three aspects will be used as a measure to evaluate students’ writing ability, see Fig. 2 for details.

Fig.2

Theoretical framework.

The assessment tool initially selected in this study was based on Jacob (Jacob) scores [23 –25], and the Jacob score table was translated into Chinese as follows, see Table 2 for details.

Table 2

Jacobs composition scale (Chinese version)

Evaluation elements and evaluation criteria	Score
Content
Intellectual, substantive, critical thinking, very persuasive.	20-17
There are some knowledge about the subject, arguments related to the article, but lack some detailed expression and some critical thinking ability.	16-13
Subject knowledge is limited, substance is limited, theme is not fully developed, debate is full of loopholes.	12-9
No knowledge was provided on the subject and was not substantive, relevant or sufficient for assessment.	8-5
Unreasonable organizational structure
Attractive words, background information and arguments should be used at the beginning; the main paragraph should contain an example of topicality or rebuttal of the opponent’s position..	20-17
According to the standard English drafting guidelines, the overall structure should include the introduction, the main paragraph, the rebuttal paragraph and the final paragraph, although some of the contents of the organizational structure are inappropriate or imperfect.	16-13
There are many organizational errors.	12-9
The organizational structure of the text is completely out of line with the English writing guidelines.	8-5
Language
Use complex sentences effectively in the text.	20-17
Use simple phrases effectively in the text.	16-13
The main problem is the separation of some sentences and the absence of components of sentences.	12-9
There are not many words in the text, the rules of structure are meaningless, there are mistakes in the whole text, readers and authors have no resonance, and there are many mistakes.	8-5
Grammar
In using consistency (quantity, sex, word or person), time, order, terms, pronouns, and terms, in a very good understanding of basic rules.There may be a small error in grammar, but the problem is small.	20-17
There are some mistakes in using consistency (quantity, sex, language or person), time, number, order, article, pronoun, etc.	16-13
Time, numbers, order, articles, pronouns, media and other words are often used incorrectly in the text.	12-9
There are a lot of errors in language grammar.	8-5
Technical level
There are very few spelling mistakes in the text.	20-17
Sometimes spelling errors, size and letter errors and interpretation ambiguities occur.	16-13
There are usually spelling errors and size errors, and the text is not good.	12-9
The text is full of spelling errors and size errors, there are many errors, the font is blurred, it is difficult to distinguish.	8-5
Total

Final evaluation tools, see Table 3–5 for details.:

Table 3

Final students’ IELTS writing ability assessment tool

Evaluation issues	Evaluation criteria
1. Topic knowledge	L4:article completely contains the topic knowledge;
	L3:article part contains topic knowledge;
	L2:articles contain limited knowledge of topics;
	L1:article does not contain topic knowledge
2. Content arguments	L4:the thesis is comprehensive;
	L3:article argument is limited;
	L2:the argument of the article is inappropriate;
	L1:the argument of the article
3. Substantive content	L4:article has substantive content;
	L3:articles have substance but lack detail;
	L2:article substantive content is less;
	L1:article has no substance
4. Arguments	L4:article shows good critical thinking;
	L3:article shows some critical thinking;
	L2:article argument has the flaw;
	L1:the argument of the article is not relevant
5. Introduction	L4:the introduction of the article has sentences, backgrounds and arguments that appeal to readers;
	L3:the introduction of the article is a little attractive, but not enough;
	L2:introduction to the article has no appeal;
	L1:didn’t see the introduction
6. Countervailing Contents	L4:article can give examples to refute each other’s position;
	L3:the article can give examples to refute each other’s position, but the examples are not appropriate;
	L2:the article can give examples to refute each other’s position, but the examples are very inappropriate;
	L1:article does not give examples to refute the other side’s position;
7. Conclusions	L4:article has strong conclusion;
	L3:the conclusion of the article is not strong enough;
	L2:conclusion of the article is unclear;
	L1:did not see the conclusion of the article;

Table 4

Language use

Evaluation issues	Evaluation criteria
8. Vocabulary Selection	L4:precise wording of the article;
	L3:article is slightly imprecise;
	L2:the article is imprecise;
	L1:article is crude
9. Simple sentence	L4:use simple sentences effectively;
	L3:article individual simple sentence has the small question;
	L2:there are some problems in the use of simple sentences;
	L1:simple sentences used in this article are missing elements and incomprehensible
10. Complex sentences	L4:use complex sentences effectively;
	L3:article individual complex sentence has the small question;
	L2:there are some problems in the use of complex sentences;
	L1:complex sentences used in this article have the problem of missing components and incomprehensible
11. Preface	L4:article word order is completely correct;
	L3:there are some mistakes in word order;
	L2:article word order often appears the mistake;
	L1:wrong word order
Articles, pronouns, prepositions	L4:articles use articles, pronouns and prepositions correctly;
	L3:articles use articles, pronouns and prepositions with individual errors
	L2:articles often make mistakes when using articles, pronouns and prepositions
	L1:article is full of errors in the use of articles, pronouns and prepositions

Table 5

Language use

Evaluation issues	Evaluation criteria
13. Number, case or name	L4:article maintains the consistency of number, case or person;
	L3:there are minor errors in maintaining the consistency of numbers, sex, case, or person;
	L2:article maintains the number, the sex, the lattice or the person to call the consistency aspect error is quite many;
14. Tenses	L4:article is completely correct in the use of tense;
	L3:there are some mistakes in the use of tense;
	L2:articles often make mistakes in the use of tense;
	L1:passage is full of mistakes in the use of tense
15. Punctuation marks	L4:article has no punctuation error;
	L3:articles occasionally have punctuation errors;
	L2:articles often have punctuation errors;
	L1:articles are full of punctuation errors

4 Theoretical basis of 4-machine learning and neural network algorithm

4.1 Circulating neural networks

Sequence data modeling often requires a balance between background features of sequence data, cyclic neural networks, and simple feedback networks, because cyclic neural networks have important advantages in designing output sequences and have an impact on current outputs, see Fig. 3 for details.

Fig.3

Circulating neural network structure and expansion.

As shown in Fig. 1, the traditional cyclic neural network expands over time, and the hidden layer is correlated by time series. the current hidden layer data comes from the current input and the current hidden layer, as described in 1: $s_{t} = σ ({Ws}_{t - 1} + {Ux}_{t})$ (1)

For gradient reverse conduction, the BPTT algorithm is used, as shown in formula (2): $\frac{\partial L_{t}}{\partial W} = \sum_{k} \frac{\partial L_{t}}{\partial s_{t}} \frac{\partial s_{t}}{\partial s_{k}} \frac{\partial s_{k}}{\partial W}$ (2) Of which $\frac{\partial s_{t}}{\partial s_{k}} = \prod_{i = 1}^{t} \frac{\partial s_{i}}{\partial s_{i - 1}} \prod_{i = 1}^{t} W^{T} diag (σ^{'} (x^{i - 1}))$ (3)

However, when the traditional network processes sequence data, the length of the sequence is often limited because the gradient is lost or exploded when the length of the sequence is longer, so it is difficult to train [26 –28]. When updating the weights (i.e., calculating the gradient), it tends to update the gradient in the direction of numerical change at the end of the order, as shown in Table 6, see Table 6 for details.

Table 6

Forward derivation algorithm LSTM

Input	Data input xtjLSTM is the network model
Output	cell unit output btc
FortintimeT:
Inputgate:	$\begin{matrix} a_{i}^{t} = \sum_{i = 1}^{I} w_{il} x_{i}^{t} + \sum_{h = 1}^{H} w_{hl} b_{h}^{t - 1} + \sum_{c = 1}^{C} w_{cl} s_{c}^{t - 1} \\ b_{l}^{t} = f (a_{l}^{t}) (4) \end{matrix}$
Forgetgate:	$\begin{matrix} a_{φ}^{t} = \sum_{i = 1}^{I} w_{i φ} x_{i}^{t} + \sum_{h = 1}^{H} w_{h φ} b_{h}^{t - 1} + \sum_{c = 1}^{C} w_{c φ} s_{c}^{t - 1} \\ b_{φ}^{t} = f (a_{φ}^{t}) (5) \end{matrix}$
Cell:	$\begin{matrix} a_{c}^{t} = \sum_{i = 1}^{I} w_{ic} x_{i}^{t} + \sum_{h = 1}^{H} w_{hc} b_{h}^{t - 1} \\ s_{c}^{t} = b_{φ}^{t} s_{c}^{t - 1} + b_{l}^{t} g (a_{c}^{t}) (6) \end{matrix}$
Outputgate:	$\begin{matrix} a_{ω}^{t} = \sum_{i = 1}^{I} w_{i ω} x_{i}^{t} + \sum_{h = 1}^{H} w_{h ω} b_{h}^{t - 1} + \sum_{c = 1}^{C} w_{c ω} s_{c}^{t - 1} \\ b_{ω}^{t} = f (a_{ω}^{t}) (7) \end{matrix}$
Celloutput:	$b_{c}^{t} = b_{ω}^{t} h (s_{c}^{t})$ (8)

4.2 Convolutional neural networks

In a convection network, the main activity is convection operation. in this case, the main function is feature extraction network transmission [29]. the convolution operation shown in Fig. 2, where the input data is the determined core coding obtained in the output section, see Fig. 4 for details.

Fig.4

Schematic diagram of convolution operation.

Pooling operations are shown in Fig. 5, see Fig. 5 for details.

Fig.5

Schematic illustration of pooling operations.

As shown in Fig. 5, the maximum value of a part of the characteristic is selected as the left seal unit size as a seal output; for the average seal, the average value is used as the seal. The sealing operation shows multidimensional features, and at the same time, the output is added. After the collection, a feature bit synthesizes most of the characteristic information close to the collection.

4.3 Residual network

In deep learning patterns, neural networks are layered, meaning that can be extracted from different features. The deeper the network, the more abstract the speech of information. however, as the depth of the network increases, the difficulty increases. At the same time, the calculation of gradient may lead to the diffusion or explosion of the network due to the additional channels entering the network depth. if the surface network is saturated, simply increasing the depth of the network can lead to its degradation. in other words, the new network component needs only a constant, and at this time the depth network is degenerating into a surface network.

Assuming that the initial calculation between the network layers is in Mode 9:

$F = W_{2} f_{1} (W_{1} x)$ (9)

After introducing the residual network, this part of the calculation is shown in formula (10). The $F = W_{2} f_{1} (W_{1} x) + x$ (10)

The introduction of a network can increase the depth of the neural network without degrading into a network surface while allowing the transmission of information through the deep layer. Therefore, during the training process, the network is described as a combination of several low-depth networks. As shown in Fig. 6, see Fig. 6 for details.

Fig.6

Of residual network.

Seq2Seq model first encodes the input sequence, and the result is the whole semantic information. Then this part of semantic information is decoded. When encoding, the neural network of the Encode part obtains the semantic vector of the sequence text C, as shown in formulas (11) and (12). The $h_{t} = f (x_{t 1} h_{t - 1})$ (11) $C = θ ({h_{1}, h_{2}, . . ., h_{m}}$ (12)

During decoding, the output Y of the current moment_tDepends on previous Y of output_t - 1 S with the overall semantic information C and the previous hidden layer state_t - 1as shown in formulas 13 and 14 $s_{t} = f (y_{t - 1}, s_{t - 1}, C)$ (13) $p (y_{t} | y_{< t}, X) = g (y_{t - 1}, s_{t}, C)$ (14)

A Attention mechanism can be used to weigh the semantic information against the weight of the current input in order to obtain semantic information with different weights for different inputs, as shown in formula (15): $s_{t} = f (s_{t - 1}, y_{t - 1}, C_{t})$ (15)

Among them, c_t The calculations are shown in formulas 16-17. $C_{t} = \sum_{j = 1}^{T} a_{tj} h_{j}$ (16) $a_{tj} = \frac{exp (e_{tj})}{Σ exp (e_{ttk})}$ (17) $e_{tj} = g (s_{t - 1}, h_{j})$ (18)

Adam optimization algorithm is an adaptive method to calculate the learning rate of each parameter. Adam optimization algorithm saves the mean value of the gradient square at the previous time and the mean value of the previous gradient, which is more robust when the gradient is updated. Its algorithm is shown in Table 7, see Table 7 for details.

Table 7

Adam algorithms

Input	Gradient gt
Output		Update gradient θ t +1
m_t = β₁m_t-1 + (1 - β₁) g_t (19)		// Cumulative gradient mean
$v_{t} = β_{2} v_{t - 1} + (1 - β_{2}) g_{t}^{2}$ (20)		// Cumulative gradient squared mean
$\overset{\land}{m_{t}} = \frac{m_{t}}{1 - β_{1}^{t}}$ (21)		// Value attenuation
$\overset{\land}{v_{t}} = \frac{v_{t}}{1 - β_{2}^{t}}$ (22)		// Value attenuation
$θ_{t + 1} = θ_{t} - \frac{η}{\sqrt{\overset{\land}{v_{t}}} + ɛ} \overset{\land}{m_{t}}$ (23)		// Final update gradient

Among them, β attenuation parameters₁Generally set to 0.9, attenuation parameter β₂ Generally set to 0.999.

5 Model design and empirical analysis of students’ IELTS writing ability index

5.1 Model design

A new measurement model was proposed by the Danish mathematician GeorgRasch in 1960 to measure the underlying psychological characteristics of the subjects. The so-called latent psychological trait is the construction that the test wants to measure. It can be ability, cognitive level, personality trait, attitude, motivation, interest, values and so on. Rasch model is defined as: ${odds}_{ni} = \frac{P_{ni 1}}{P_{ni 0}} = \frac{θ_{n}^{'}}{δ_{i}^{'}}$ (24)

The ratio of the ability of the subjects is independent of the characteristics of the test questions, which is called objectivity. Taking the scale as the natural base logarithm, we can get

$\begin{array}{l} \log i t_{n i} & = \log (o d s s_{n i}) = \log (\frac{θ_{n}^{'}}{δ_{i}^{'}}) \\ = \log (θ_{n}^{'}) - \log (δ_{i}^{'}) = θ_{n} - δ_{i} \end{array}$ (25)

The measurement results of the two subjects were compared and obtained. $log {it}_{1 i} - log {it}_{2 i} = (θ_{1} - δ_{i}) - (θ_{2} - δ_{i}) = θ_{1} - θ_{2}$ (26)

Obviously, the difference of the potential characteristics of the subjects was independent of the characteristics of the test questions. hence Rasch model is an objective scale. However, the value of the two subjects will not change, so the Raschlogit model is also an equidistant scale.

RASCH model has both descriptive and predictive functions.

As for the description, it can clearly explain the relationship between the possible psychological characteristics of the subject and the characteristics of the evidence problem, the differences between the subjects and the psychological problems of the subject. in the Rasch model, the difficulty of testing the problem lies in the probability of the answer. by observing the logical value of the subject, the difference between the two can be clearly explained.

In terms of prediction, it is possible to predict whether candidates with certain abilities are likely to answer a difficult question correctly. The difference between the potential psychophysiological characteristics of the R candidate and the characteristics of the test item is 0.5. If the basic psychological characteristics exceed the characteristics of the test item, the probability of reply exceeds 0.5; when the potential psychological characteristics are lower than the characteristics of the test, the probability of response is less than 0.5.

The evaluation index in the scorecard is an important part of measuring the quality of the evaluation table, Through indicators of consistency, It can identify indicators on the evaluation list that are problematic, The question is what and how they can be modified. Tables 3 to 14 present the proposed indicators for the 15 assessment indicators in the scale. In general, On the basis of the MSQ and ZSTD values in the INFIT and OutFIT proposed adjustment indices, MSQs values between 0.70 and 1 are acceptable. According to this standard, As you can see, Assessment indicators 13 and 17 are above 1.3, ZSTD values exceed 2 per cent; Assessment of ZSTD values for indicators 8,12 and 17 exceeding 2; INFIT values are below –2, Therefore, These assessment indicators have some problems with data compatibility, Improvements are needed at the next review stage. In the scale of the assessment, These data largely reflect and evaluate students’ writing ability, see Table 8 and Figs. 7–9 for details.

Table 8

Fitting indices of topics in the scale

Title	Difficulty estimate	Standard error	Infit		Outfit		PTMEA value
			MNSQ	ZSTD	MNSQ	ZSTD
T 01	–0.01	0.16	0.73	–2.80	0.74	–2.36	0.81
T02	–0.13	0.15	078	–2.13	0.75	–1.91	0.81
T03	0.12	0.15	0.93	–0.55	0.98	–0.05	0.75
T04	0.92	0.14	0.72	–2.97	0.71	–2.79	0.83
T 05	0.16	0.15	0.88	–1.13	0.86	–1.07	0.78
T07	–0.51	0.16	0.85	–1.29	0.79	–1.45	0.78
T08	0.08	0.15	1.26	2.21	1.32	2.28	0.68
T09	2.98	0.15	0.76	–2.39	0.73	–2.45	0.85
T10	1.59	0.14	0.90	–0.93	0.89	–1.01	0.80
T11	0.11	0.15	0.88	–1.14	0.89	–0.76	0.79
T12	–0.82	0.19	1.21	1.64	2.46	4.23	0.62
T13	–2.35	0.18	1.32	2.58	1.08	0.34	0.57
T14	–0.49	0.15	1.00	0.03	1.03	0.20	0.75
T15	0.73	0.13	1.21	1.75	1.37	2.39	0.75
T17	–2.36	0.18	1.59	4.14	4.39	4.38	0.43

Fig.7

Fitting indices of topics in the scale.

Fig.8

Difficulty estimate.

Fig.9

Standard error.

The difference values of the RASCH model are expressed in Logit values and are not easy to replicate and interpret. in order to facilitate users to understand the difficulty and ability of the evaluation indicators, RASCH model provides raw and RASCH values for conversion to help users estimate the tested capacity levels (see Tables 3–9). in english writing ability assessment, the lowest value is 16 points, the RASCH value is –7.38, the highest value is 60 points, and the Rasch score is 8.63 points, see Table 9 for details.

Table 9

Conversion table of raw and Rasch scores

Original score	Rasch	S.E.
16	–7.38	1.85
17	–6.11	1.04
18	–5.33	0.77
19	–4.84	0.65
20	–4.45	0.59
21	–4.12	0.55
22	–3.83	0.53
23	–3.56	0.51
24	–3.31	0.50
25	–3.07	0.49
26	–283	0.48
27	–2 60	0.48
28	–2.37	0.47
29	–2.15	0.47
30	–1.93	0.47
31	–1.72	0.46
32	–1.51	0.46
33	–1.29	0.46
34	–1.08	0.46
35	–0.87	0.46
36	–0. 66	0.46
37	–0.45	0.46
38	–0.23	0.47
39	–0.01	0.47
40	0.22	0.48
41	0.45	0.49
42	0.69	0.49
43	0.94	0.50
44	1.20	0.51
45	1.46.	0.52
46	1.73	0.52
47	2.01	0.53
48	229	0.53
49	2 57	0.53
50	2.85	0.53
51	3.14	0.54
52	3.43	0.55
53	3.74	0.56
54	4.06	0.58.
55	4.42	0.62
56	4.83	0.67
57	5.34	076
58	6.02	0.91
59	7.09	1.20
60	8.63	1.96

5.2 Empirical analysis

5.2.1 At different levels

A characteristic high quality sample that embodies the highest evaluation standard of writing ability, which requires the writer to be “clear “,” simple” and “true “. therefore, the writer of high quality sample must have a high level of english expertise, structure and consistency.

In terms of its characteristics and cause analysis, the evaluation criteria of medium-level samples should be partially in line with the English ability of English teachers. The authors of this kind of text are basically able to use a simple, correct and basic vocabulary to understand the order of the text and the use of the article, which will make individual mistakes.

There are three main reasons for poor writing: on the one hand, the author’s own logic is problematic and can not focus on writing; on the other hand, the daily accumulation of English phrases is limited and lack of grammatical skills; on the other hand, the author’s attitude is not good. English professional skills are very low.

5.2.2 Strategies for developing students’ IELTS writing ability

Examination and Conception in IELTS Writing:

According to the written requirements of IELTS, the composition of IELTS usually consists of three paragraphs, and the first paragraph should contain at least three elements: a background phrase that describes the views and phenomena of the topic, as well as opinions and positions. In the process of teaching, teachers should guide students to learn the theme, then read the theme, analyze the main information of the theme, understand the content of the theme, and understand the idea of the theme, so that students can extract and synthesize the information. Improve their thinking ability. Teachers should guide students to develop courses to avoid confusion.

Creation of arguments:

Once the position on this issue is set out in the first paragraph, arguments must be made and arguments must be clear and well documented. To cultivate students’ writing ability, teachers should generally be guided by objective and fair evaluation, explain articles or news related to IELTS topics, such as education, government, economy, culture, etc., and enable students to carry out objective evaluation.

Analysis of the conclusions:

IELTS compositions are usually divided into topics related to debates and reports, that is, debates, which require candidates to resubmit the main arguments on shorter topics and draw partial conclusions on the final purpose of the debates, propose solutions and respond to the issues involved. The topic of the report requires students to analyze the phenomenon, outline its causes, and finally make suggestions. With regard to these two different topics, teachers can teach students how to read national and international media or related social activities in normal classes, so that they can understand the drafting style of English articles and how to deal with social problems.

5.2.3 A Comprehensive teaching method for L2 writing

L2 writing is a complex process of cognition and language transformation, which transcends the simple “writing” and takes into account the factors of language foundation, reading, culture, thinking mode and so on. Reading-debating-writing must be the main form.

The breadth and depth of thought affect the creativity of thought, which can promote students’ learning and produce good learning effect. Language learning includes newspapers, magazines, Internet articles, novel excerpts, songs, speeches and so on. They better reflect the use of English in real life and make the way of reading more abundant. Reading process is a mixed learning process, which helps to motivate students, encourage students to think positively and cultivate students’ perseverance, so that students can transfer the cultivation of thinking ability in class to real life. The real goal is to develop this ability. Normative language is full of individualized thinking, language thinking can not achieve its goal, writing is a custom writing behavior, we must emphasize personal experience life. Students can be allowed to write articles of similar patterns, but its biggest drawback is that it is difficult to shape students’ thoughts.

Therefore, in IELTS teaching, combining teaching methods can make students adapt to the requirements of the examination more quickly. Second language writing teachers need to understand the different writing traditions of other cultures. And work together to understand the different writing methods of different cultures in writing. Teachers and students are urged to communicate with each other and to gain universal recognition in different fields as far as possible. It emphasizes the communication and discussion between teachers and students, and the communication between students is also an effective part of writing practice. Cooperative learning provides students with the opportunity to listen to different opinions and discover the strengths of others. It helps to create responsive ways of thinking.

The ability to write not only refers to the ability to write, but also refers to the ability to change, judge, learn lessons, and so on, which can change the single model of “teacher assessment, student perspective “, while paying attention to the interaction and diversity of practices among the subjects under assessment, and using successive individual review cycles to modify the text to truly reflect and strengthen the universal consciousness of the organizations. Students should recognize the importance of cooperation and benefit from self-assessment. Under the changing background, new ideas are being formed, and students are beginning to realize that they should avoid low-level writing mistakes, pay attention to the training of logic and critical thinking, and improve the level and quality of thinking at the same time. On the other hand, teachers can provide more specific and effective teaching.

6 Conclusion

Based on the study of machine learning algorithms and neural networks, this study selected and revised English writing ability assessment tools. Based on experience, the reliability and utility of the tool are tested. By using the SPSS21.0 plural statistical tool, the writing ability of IELTS students is evaluated, and the quality test is carried out by using the model. The results show that the overall quality of the evaluation tool is verified when analyzing the rationality of the evaluation index. The research shows that the evaluation index of students’ English writing ability reflects and evaluates their English writing ability to a great extent. The research shows that students’ English writing ability is generally very high, most of their thoughts and expressions meet the relevant standards, but there is a big gap in the use of language. In general, the use of evaluation tools can assess students’ English writing ability, and more specific and effective.

References

Mistry

, Zhang

, Neoh

S.C.

, Lim

C.P.

and Fielding

, A micro-GA embedded PSO feature selection approach to intelligent facial emotion recognition, IEEE Trans Cybern 47(5) (2016), 1496–1509.

Mitra

and Acharya

, Gesture recognition: a survey, IEEE Trans Syst Man Cybern Part C 37(3) (2007), 311–324.

Mollahosseini

, Chan

and Mahoor

M.H.

, Going deeper in facial expression recognition using deep neural networks, In: IEEE winter conference on applications of computer vision (WACV), Lake Placid, NY 11(3) (2016), 1–10.

Nie

X.H.

, Wang

and Nie

H.Y.

, Chaos quantum-behaved Cat Swarm Optimization Algorithm and its application in the PV MPPT, Comput Intell Neurosci 31(4) (2017), 89–102.

Peng

, Xia

, Li

and Feng

, Towards facial expression recognition in the wild: a new database and deep recognition system, In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR) workshops 26(6) (2016), 27–30.

Peters

, Janzing

and Schlkopf

, Identifying cause and effect on discrete data using additive noise models, In: Proceedings 13th international conference artificial intelligence and statistics 10(9) (2010), 597–604.

Ranjan

, Patel

V.M.

and Chellappa

, HyperFace: A deep multitask learning framework for face detection, landmark localization, pose estimation, and gender recognition, IEEE Trans Pattern Anal Mach Intell 41(1) (2019), 121–135.

Rehab

A.L.

, Abd Elaziz

and Lu

, Chaotic opposition-based grey-wolf optimization algorithm based on differential evolution and disruption operator for global optimization, Expert Syst Appl 108(1) (2018), 1–27.

Samara

, Galway

, Bond

and Wang

, Affective state detection via facial expression analysis within a human–computer interaction context, J Ambient Intell Human Comput 10(8) (2019), 2175–2184.

10.

Shan

, Gong

and Mc Owan

P.W.

, Facial expression recognition based on local binary patterns: a comprehensive study, Image Vis Comput 227(6) (2009), 803–816.

11.

Shu

C.C.

and Tsai

F.W.

, Computational intelligence based on the behavior of Cats, Int J Innov Comput Inf Control 3(1) (2007), 163–173.

12.

Song

, Kim

, Jeon

P.B.

, Deep learning for real-time robust facial expression recognition on a smartphone, In: IEEE international conference on consumer electronics (ICCE), Las Vegas, NV 53(2) (2014), 564–567.

13.

Sreevatsan

A.N.

, Sathish Kumar

K.G.

, Rakeshsharma

and Roomi

M.M.

, Emotion recognition from facial expressions: a target oriented approach using neural network, In: Proceedings of the indian conference on computer vision, graphics and image processing 28(4) (2004), 1–6.

14.

Tai

S.C.

and Chung

K.C.

, Automatic facial expression recognition using neural network, In: IEEE Region 10 Conference—TENCON 25(3) (2007), 689–699.

15.

Walecki

R.R.

and Deep

, structured learning for facial expression intensity estimation, Image Vis Comput 25(9) (2017), 143–154.

16.

Wang

, Otto

and Jain

, Face search at scale: 80 million gallery, Comput Res Repos 15(7) (2015), 1358–1366.

17.

Wozniak

and Połap

, Adaptive neuro-heuristic hybrid model for fruit peel defects detection, Neural Netw 98(4) (2018), 16–33.

18.

Wozniak

and Połap

, Bio-inspired methods modeled for respiratory disease detection from medical images, Swarm Evol Comput 16(2) (2018), 1016–1025.

19.

Yang

X.S.

, Engineering optimization: an introduction with metaheuristic applications, Hoboken 25(3) (2010), 356–366.

20.

Yolcu

, Oztel

, Kazan

, et al., Deep learning-based face analysis system for monitoring customer interest, J Ambient Intell Human Comput 11(6) (2020), 237–248.

21.

Zeng

, Zhang

, Song

, et al., Facial expression recognition via learning deep sparse autoencoders, Neurocomputing 27(3) (2018), 643–649.

22.

Ahmed

, Iqbal

M.M.

, Aldabbas

, Khalid

, Saleem

and Saeed

, Images data practices for Semantic segmentation of breast cancer using deep neural network, J Ambient Intell Hum Comput 65(2) (2020), 569–575.

23.

Al-Dmour

and Al-Ani

, A clustering fusion technique for MR brain tissue segmentation, Neuro Comput 27(5) (2018), 546–559.

24.

Amin

, Sharif

, Raza

and Yasmin

, Detection of brain tumor based on features fusion and machine learning, J Ambient Intell Hum Comput 19(4) (2018), 1092–1125.

25.

Angulakshmi

and Lakshmi Priya

G.G.

, Brain tumour segmentation from MRI using superpixels based spectral clustering, J King Saud Univ Comput Inf Sci 10(9) (2018), 2445–2513.

26.

Bharathi

and Satish

, Improving MRI brain tumor classification and segmentation based on comparison technique, Int J Emerg Technol Comput Sci Electron (IJETCSE) 13(1) (2015), 560–569.

27.

Borole

V.Y.

, Nimbhore

S.S.

and Kawthekar

D.S.S.

, Image processing techniques for brain tumor detection: a review, Int J Emerg Trends Technology Comput Sci (IJETTCS) 4(5) (2015), 28–32.

28.

Chanchlani

, Chaudhari

, Shewale

B.Y.

and Jha

, Tumor detection in brain MRI using clustering and segmentation algorithm, Imperial J Interdiscip Res 3(5) (2017), 2122–2127.

29.

Kaur

and Rani

, MRI brain tumor segmentation methods—a review, Infinite Study 6(3) (2016), 760–764.