Page 67 - IJOCTA-15-1

BSO: Binary Sailfish Optimization for feature selection in sentiment analysis

Table 3. Example of comments

# of sample Sample comment Class
The camera quality is excellent in terms of both price and performance;
1 Negative
however, the battery drains quickly and it overheats.
2 The shoe is small in size but very comfortable and lightweight. Positive

Figure 2. Proposed BiLSTM structure

Table 4. Determined parameter’s values for BiLSTM model

Parameter Name Search Space Best
Maximum length {150, 200, 300} 300
Batch size {16, 32, 64} 32
Epoch {5, 10, 15} 10
Vocabulary size {1000, 5000, 10000, 15000, 20000} 10000
Embedding dimension {100, 150, 200, 300} 300
Kernel unit for BiLSTM layer {16, 32, 64} 64

i
each preprocessing technique C within a com- 4.2.1. Determination of the appropriate k
j
bination belongs to the set [0, 1]. A value of 0 value
i
indicates that the preprocessing technique C is
j
i
not applied, while a value of 1 signifies that C is In this section, we determine the optimal number
j
implemented for the combination C j . Finally, we of words k to be utilized from the vocabulary V
compare the performance evaluations of the BiL- when converting comments into TF-IDF vectors.
STM model against the results obtained from the The values tested for k include 1000, 5000, 10000,
aforementioned ML algorithms. 15000, and 20000. These values represent the k

61