Emotion analysis method for character and emoji combined text

Abstract

The invention discloses a text and emoji combined text sentiment analysis method, which comprises the following steps of S1, collecting basic data, and performing primary classification preprocessing on comment short text data according to a preset rule; s2, characters and emotion symbols are split through a python program, and emotion judgment is carried out on the samples subjected to classification preprocessing; and S3, carrying out merging calculation on the judged two types of emotion results according to a preset rule, storing result data of emotion calculation meeting the rule into a database, and ending emotion judgment. According to the scheme, result data after emotion judgment is ensured to be accurate and available, the real customer emotion tendency of automobile products in the aspects of sales, service, product design, product quality and the like is objectively and truly displayed, then customer demand points are found, and the product recognition degree and public praise are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of social short text emotion determination in the automotive industry, and in particular to a method for analyzing the emotion of a text combined with words and emojis. Background Art

[0002] In the Web2.0 era, social media platforms based on mobile Internet are booming, and hundreds of millions of voices express netizens' opinions and emotional tendencies on various hot events or products in the physical world. Among the 840,611 weekly voices of two car brands counted on the two social platforms with the largest daily voices, 328,962 voices included emojis (emoticons), accounting for 39.1%, involving 342 emotional symbols. Emojis have a stronger visual impact than texts due to their intuitive and vivid characteristics and are more capable of accurately expressing emotions. However, emojis themselves have the characteristics of large number, great differences in sending frequency, a wide variety, and ambiguous emotional expression. In social short texts where text and emoji coexist heterogeneously, there are difficulties in emotional calculation and analysis.

[0003] Existing technologies lack the ability to define and calculate the sentiment of texts that are a combination of text and emojis. Some products focus on natural language understanding and sentiment determination of pure text, and directly delete emojis as noise when analyzing. Other products perform emoji sentiment definition calculation, emoji selection recommendation, and application separately, without considering the mutual influence and effect of text and emoji on sentiment.

[0004] The physical world, the information world and human society constitute the ternary world of the new information worldview. People's social needs have prompted netizens to no longer be just content receivers, but to become content providers through forwarding, liking, commenting, interacting, private messaging, etc. Social media platforms have become tools and gathering places. Massive social short texts have promoted the rapid spread of events in the physical world. These information rich in opinions and emotions provide effective data support and scientific basis for public opinion supervision decisions, corporate product improvements or advertising. Based on incomplete statistics from "EMOJI Chinese Network", there are more than 4,000 Emojis, which can be divided into 10 major categories: smiley faces & emotions, humans & bodies, skin colors & hairstyles, animals & nature, food & beverages, travel & places, activities, objects, symbols, and flags, which are subdivided into 110 subcategories. Among the 342 emoji expressions counted for the weekly voice of all models of the two automobile brands on the two major social media platforms, a total of 783,496 emojis appeared under the premise of counting the single, continuous and overlapping uses of emojis, of which the first 92 emojis accounted for 98%, satisfying the typical "long tail phenomenon". Therefore, determining the emotional tendencies of the first 92 emojis can cover most of the usage scenarios of emojis in social short texts in the automotive industry. Summary of the invention

[0005] Based on the above problems, the present invention proposes a sentiment analysis method for text combined with text and emoji, and the specific technical solution is as follows.

[0006] A sentiment analysis method for text combined with emojis, comprising the following steps:

[0007] Step S1: Collect basic data and perform classification preprocessing on the comment short text data according to pre-set rules;

[0008] Step S2: Split the text and sentiment symbols through the Python program, and make sentiment judgment on the samples after classification preprocessing;

[0009] Step S3: The two types of emotion results are combined and calculated according to the pre-set rules, and the result data of the emotion calculation that meets the rules is stored in the database, and the emotion judgment is completed.

[0010] Specifically, the collection of the basic data includes:

[0011] The short texts of netizens’ comments on two major social media platforms were collected by crawler program according to car model keywords;

[0012] The newly collected text, author and publication time fields are combined to generate a 16-bit unique value RowKey, which is compared with the existing RowKey in the original database. If there is a duplicate, the processing flow is terminated; if there is no duplicate, the next step is executed.

[0013] Specifically, the classification preprocessing includes:

[0014] Pre-process and filter the text according to the set standards, and mark the short texts containing both text and emoji as the corpus for sentiment judgment;

[0015] Store samples that meet the preprocessing conditions into the specified database, and clean up the corpus database regularly;

[0016] The samples that do not meet the text and emoji combination preprocessing conditions are transferred to other processing branches.

[0017] Specifically, the emotion determination of the sample includes:

[0018] The text in the sample after classification preprocessing is subjected to sentiment determination according to the pre-trained sentiment determination algorithm;

[0019] The emoji in the sample after classification preprocessing are subjected to sentiment judgment according to the manually set sentiment values.

[0020] Specifically, the results of the emotion determination include: positive, neutral and negative.

[0021] Specifically, the emoji emotion judgment also includes single expression and multiple expressions.

[0022] Beneficial effects of the present invention: The present invention proposes a sentiment analysis method for text combined with text and emojis. Based on the accumulation of sentiment analysis case experience of massive automotive industry Internet user voice data and the precipitation of long-term manually labeled data, a set of scientific sentiment judgment methods are produced to ensure that the result data after sentiment judgment is accurate and available, and objectively and truly show the real emotional tendencies of customers in sales, service, product design, product quality and other aspects of automotive products, thereby finding customer demand points and improving product recognition and reputation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of emotion determination rules for the combination of text and emoji in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0025] The present invention proposes a sentiment analysis method for texts combined with emojis. After writing the collected raw data into a specified database, a text classification preprocessing is performed to identify short texts containing both text and emojis as a corpus for sentiment determination. Next, sentiment determination is performed on the text part and the emoji part respectively. The two types of sentiment results are then combined and calculated, and the result data of the sentiment calculation is stored in a database. Finally, a manual periodic summary is performed, and the migration results of the emoji category library and the sentiment value are further updated through a large number of sample annotations and data calculations to adapt to the high-frequency changes of social media and the emergence of new words on the Internet.

[0026] In a preferred embodiment, the sentiment analysis method of text combined with emoji includes the following steps:

[0027] Step S1: Collect basic data and perform classification preprocessing on the comment short text data according to pre-set rules;

[0028] Step S2: Split the text and sentiment symbols through the Python program, and make sentiment judgment on the samples after classification preprocessing;

[0029] Step S3: The two types of emotion results are combined and calculated according to the pre-set rules, and the result data of the emotion calculation that meets the rules is stored in the database, and the emotion judgment is completed.

[0030] In this embodiment, the collection of basic data includes:

[0031] The short texts of netizens’ comments on two major social media platforms were collected by crawler program according to car model keywords;

[0032] The newly collected text, author and publication time fields are combined to generate a 16-bit unique value RowKey, which is compared with the existing RowKey in the original database. If there is a duplicate, the processing flow is terminated; if there is no duplicate, the next step is executed.

[0033] In this embodiment, classification preprocessing includes:

[0034] Pre-process and filter the text according to the set standards, and mark the short texts containing both text and emoji as the corpus for sentiment judgment;

[0035] Store samples that meet the preprocessing conditions into the specified database, and clean up the corpus database regularly;

[0036] The samples that do not meet the text and emoji combination preprocessing conditions are transferred to other processing branches.

[0037] In this embodiment, the emotion determination of the sample includes:

[0038] The text in the sample after classification preprocessing is subjected to sentiment determination according to the pre-trained sentiment determination algorithm;

[0039] The emoji in the sample after classification preprocessing are subjected to sentiment judgment according to the manually set sentiment values.

[0040] In this embodiment, the results of emotion judgment include: positive, neutral and negative. The emotion dimension observation theory model divides human emotions into pleasure (P), activation (A) and dominance (D), where P represents the positive or negative nature of the subject's emotional state, that is, the degree of positivity or negativity of the emotion, the degree of like or dislike, and this dimension reflects the essence of emotion. Usually, when doing emotion calculation, it is divided into three categories: positive, neutral and negative.

[0041] In order to effectively calculate the sentiment of a text combined with emojis, we made two assumptions: 1. Assume that the sentiment value of a social text is unique (that is, the sentiment value is one of positive, neutral, and negative); 2. The sentiment of a single or multiple emojis in the same social short text is consistent with the sentiment expressed by the text.

[0042] In this embodiment, emoji emotion determination also includes single expression and multiple expressions.

[0043] In another embodiment, in view of the lack of definition and calculation of text emotions of the combination of text and emoji in the prior art, some products focus on the natural language understanding and emotion judgment of pure text, and ignore emoji as interference noise. When the meaning of the text expression contains metaphors, irony, sarcasm, and implicit meaning, the ignored emoticons play a decisive role in the real emotion expressed in the short text. Another part of the products perform emoji emotion definition calculation, emoji selection recommendation and application separately, without considering the impact of text on emoji emotion migration as social situations continue to evolve, resulting in deviations or errors in the judgment of emotional tendency. The sentiment analysis method proposed in the present invention includes the following steps:

[0044] S1, first collect basic data:

[0045] S1.1, collect short texts of netizens’ comments on two major social media platforms by car model keywords through a crawler program.

[0046] S1.2, combine the newly collected text, author, and publication time fields to generate a 16-bit unique value RowKey, and compare it with the existing RowKey in the original database. If there is a duplicate, the processing flow is terminated, otherwise proceed to the next step.

[0047] S2, perform classification preprocessing on the comment short text data according to the pre-set rules:

[0048] S2.1, pre-process and filter the text according to the set standards, and mark the short texts containing both text and emoji as the corpus for sentiment judgment;

[0049] S2.2, samples that meet the preprocessing conditions are stored in the designated database. The corpus database needs to be cleaned up regularly to improve the processing speed;

[0050] S2.3, transfer the samples that do not meet the pre-processing conditions of the combination of text and emoji to other processing branches. In this embodiment, the pure text samples are transferred to the emotion judgment branch, and emotion judgment is performed according to the pre-trained emotion judgment algorithm. The emotion judgment algorithm can adopt a variety of calculation methods according to needs.

[0051] S3, splits text and sentiment symbols through Python program.

[0052] S4, performing sentiment determination on the text in the sample after classification preprocessing according to the pre-trained sentiment determination algorithm.

[0053] S5, the emoji in the sample after classification preprocessing is subjected to sentiment judgment according to the manually set sentiment value, such as Figure 1 .

[0054] S6, the two types of emotion results are combined and calculated according to the pre-set rules, and the result data of the emotion calculation that meets the rules is stored in the database, and the emotion judgment is completed; for the samples that do not meet the rules, the next step is executed.

[0055] S7, manually review and label samples with inconsistent emotions identified by the rules.

[0056] S8, the final periodic summary, updates the emoji category library and the migration results of sentiment values ​​through a large number of sample annotations and data calculations to adapt to the high-frequency changes in social media and the emergence of new Internet words.

[0057] By matching preset rules, we can perform sentiment analysis and calculation on texts that are a combination of text and emojis. We collect massive data from two major social media platforms, combine big data calculations with manual annotation, and develop a scientific sentiment judgment model to ensure that the result data after sentiment judgment is accurate and available. This has become a powerful tool for enterprise product design and optimization, as well as pre-sales and after-sales service improvement.

[0058] In this embodiment, the Emoji emotion determination rules are as shown in Table 1:

[0059] Table 1 Emoji emotion judgment rules

[0060]

[0061]

[0062]

[0063] The present invention has shown and described the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A sentiment analysis method for text combined with emoji, characterized in that: The following steps are involved: Step S1: Collect basic data and perform classification preprocessing on the comment short text data according to pre-set rules; Step S2: Split the text and sentiment symbols through the Python program, and make sentiment judgment on the samples after classification preprocessing; Step S3: The two types of emotion results are combined and calculated according to the pre-set rules, and the result data of the emotion calculation that meets the rules is stored in the database, and the emotion judgment is completed.

2. The method for sentiment analysis of a text combined with emojis according to claim 1, characterized in that: The collection of basic data includes: The short texts of netizens’ comments on two major social media platforms were collected by crawler program according to car model keywords; The newly collected text, author and publication time fields are combined to generate a 16-bit unique value RowKey, which is compared with the existing RowKey in the original database. If there is a duplicate, the processing flow is terminated; if there is no duplicate, the next step is executed.

3. The sentiment analysis method of a text combined with emoji according to claim 1, characterized in that: The classification preprocessing includes: Pre-process and filter the text according to the set standards, and mark the short texts containing both text and emoji as the corpus for sentiment judgment; Store samples that meet the preprocessing conditions into the specified database, and clean up the corpus database regularly; The samples that do not meet the text and emoji combination preprocessing conditions are transferred to other processing branches.

4. The method for sentiment analysis of a text combined with emojis according to claim 1, characterized in that: The emotion determination of the sample includes: The text in the sample after classification preprocessing is subjected to sentiment determination according to the pre-trained sentiment determination algorithm; The emoji in the sample after classification preprocessing are subjected to sentiment judgment according to the manually set sentiment values.

5. The method for sentiment analysis of text combined with emoji according to claim 4, characterized in that: The results of the emotion determination include: positive, neutral and negative.

6. The method for sentiment analysis of text combined with emoji according to claim 4, characterized in that: The emoji emotion judgment also includes single expression and multiple expressions.

Images