Exploring the influence from media reports to the impression of 7-ELEVEn with neural networks (以類神經網路探查媒體報導對統一超商印象造成的影響)

In this test case, we conducted two tests, including changing the method of calculating the semantic vector from "addition" to "subtraction", and testing the impact that online media reports may have on brand impressions. The main reason why the semantic vector is adjusted to "subtract" is because in other comparison tests, it is found that such calculation output is more similar to human thinking. The establishment of branding logic uses the words "KFC" and "Fried chicken" which are more direct and simple. So the logic calculation method can be generated as follows:
在這一次的測試範例中,我們進行兩項測試,包括將語意向量的計算方法從「加法」改為「減法」,以及測試網路媒體報導可能對品牌印象造成的引響。語意向量調整為「減法」的主要原因,是因為在其他的對比測試中,發覺這樣的計算輸出結果與人為思考更為相近。品牌印象邏輯的建立,則採用印象較直接且單純的「肯德基」與「炸雞」這兩個詞彙。因此可產生邏輯計算方法如下:

"KFC" - "Fried chicken" = "小七, nickname of 7-ELEVEn" - "What's the corresponding impression?":
「肯德基」-「炸雞」=「小七」-「有什麼對應的印象?」

In this formula, we also test vocabulary such as "統一", "Superstore" and "Convenience store". Among them, the results of "小七, nickname of 7-ELEVEn" are best. The reason for this is that the term "統一" is also widely used in political news. "Supermarket" and "Convenience store" are seriously confused with the impressions of other brands of convenience store. "小七" is a specific term for 7-ELEVEn.
在這個公式中,我們也測試「統一」、「超商」與「便利商店」等詞彙,然而其中以「小七」的結果最好。其原因在於「統一」這詞彙也廣泛出現在政治類新聞,「超商」與「便利商店」則與其他品牌便利商店的印象嚴重混和,「小七」則是針對7-ELEVEn的專一詞彙。

Among the 200 vocabularies with the highest vector similarity, 195 meaningful vocabularies can be obtained through human judgment (the accuracy of the word segmentation is 97.5%). Then it can be sorted and classified according to the characteristics of the vocabulary:
在向量相似度最高的 200個詞彙中,人為判斷可獲得 195個有意義詞彙(斷詞正確率為 97.5%)。然後可以依據詞彙所屬性質,排序與分類如下:

 

Vocabulary type
詞彙類型
Vocabulary content
詞彙內容
Food vocabulary
食品類詞彙
飯糰、關東煮、霜淇淋、三明治、雪糕、雞蛋糕、肉圓、飯團、烤雞、紅豆餅、刨冰、鬆餅、炸物、可頌、章魚燒、鮮食、冰棒、小火鍋、土司、雞翅、布丁、餃子、吐司、貝果、麻辣燙、燴飯、冰沙、零食、咖哩、甜點、蛋餅、鮮果、可麗餅、冰淇淋、辣雞球、黑糖、珍珠、火腿蛋、鹹酥雞、熱狗、香雞排、蒸餃、豬排、餐包、蛋塔、奶酥、蔥油餅、甜品、豆乳、肉蛋、貝禮詩、舒芙蕾、燒餅、豆花、炸肉、果茶、水餃、酷繽沙、綠豆沙、司康、冰品、酥餅、紅茶、脆片、甜筒、生菜沙拉、雙醬、哈密瓜、肉鬆、蛋餃、芋泥、糕點、玉米濃湯、薯餅、焙茶、玉子燒、年糕、養樂多、飲料、比薩、茶葉蛋、鍋貼、甜湯、洋芋片、潤餅、肉乾、麻婆豆腐、餡餅、雞胸、餅乾、春捲、烤肉串、番薯、披薩、蛋撻、起士、生乳、麵、菠蘿、蔥肉、蛋包飯、芝麻湯圓、生啤酒、夾心、花生醬、串燒、冬瓜茶、燒肉、曲奇、海苔、飲品、沙拉、雞塊、豆沙、麥茶、水煎包、哈根達斯、奶茶、蒸飯、熱食、鵝肉、饅頭、壽喜燒、點心、輕食、貢丸湯、糰子、牛舌餅、雞球、薄餅、炸雞腿、雞汁、冷飲、雞腿飯、簡餐、豬腸、肉包、仙草
Industry vocabulary
產業類詞彙
店中店、便利商店、超商、萊爾富、冰店、光泉、全聯、店、寶雅、烤肉店、全家、711、小店、專門店、飲料店、連鎖店、subway、特色店、便利店、lawson、麵屋一燈、頂好、eleven、專賣店、港點
Description vocabulary
描述類詞彙
現烤、韓式、冰心、脆皮、現煮、口味、現炸、現做、現點、爆餡、現包、炭烤、辣味、原味、熱壓、單賣、大罐、涼、獨賣、麥香、椰香、青醬、特調、蒜味、總匯、嚐鮮、超甜
Other vocabulary
其他類詞彙
嚕嚕米、我都點、玻璃罐、史努比、貓掌
Wrong vocabulary
錯誤斷詞
店還、超商賣、5903、店有、店中


As the information in this table is based on the "KFC" to compare the vector correlation of "Fried chicken" to find out the words that have similar relevance to "小七, nickname of 7-ELEVEn". Therefore, the use of the term "Fried chicken" has made most of the vocabulary found highly relevant to food. In addition, the neural network algorithm will also find out the industry category vocabulary that appears in articles with same characteristics. There are vocabularies for convenience stores in the industry category vocabulary and brand vocabulary related to 7-ELEVEn. And most of the description vocabulary is related to the way food is made or tasted. Lastly, in other vocabulary words, the name of the endorsed cartoon character of the brand cooperation appears.
由於這個表格中的資訊,是依據「肯德基」對「炸雞」的向量相關性去比對數據,找出對「小七」具有類似相關性的詞彙。因此「炸雞」一詞的使用,使大部分找出的詞彙與食品有高度關聯。此外,類神經網路演算法亦將在同樣性質文章出現的產業類別詞彙也找出來。其中有關於便利商店相關的詞彙描述,也有與 7-ELEVEn相關的產業品牌。而描述類的詞彙中,則多數與食品的製作方式或口味相關。最後在其他類詞彙中,則出現品牌合作的代言卡通人物名稱。

In summary, this analysis method can be used to evaluate the final results of a large number of news reports about 7-ELEVEn. Since the neural network vector analysis model is based on the observation of the signal transmission of neuronal physiology, we boldly assume that the lexical relevance calculated by the neural network will be similar to the impression produced by consumer reading news or public opinion. So we can use this calculation method to assess whether media reports, marketing operations, or ad placement have the opportunity to produce the desired results.
總結來說,這樣的分析方法可以用於評估大量關於 7-ELEVEn的新聞報導的最終產生結果。由於類神經網路的向量分析模型是建立於神經細胞生理的訊息傳導觀察,因此我們大膽假設類神經網路計算出的詞彙關聯性,會與消費者閱讀新聞或輿論報導所產生的印象相似。於是我們可以藉由這樣的計算方法,評估媒體報導、行銷操作或廣告投放是否有機會產生理想中的結果。

Finally, welcome to contact us, if you have any needs about large-scale data analysis .
最後,如果您有領域相關或任何大規模資料分析的需求,歡迎與我們聯繫。