r - 在 R 中使用 LDA 按组进行主题建模

Question

我被困在一个问题上。我正在尝试使用 LDA 将句子分类为主题。我已经做到了，但问题是：LDA 正在处理整个数据集，并为我提供整个数据集的主题术语。我想在数据集中按组获取主题术语。

所以我的数据看起来像这样：

Comment                                                                                  Division
Smooth execution of Regional Administration in my absence. Well done.                    Finance
Job well done in completing CPs and making the facility available well in time.          Finance
Good Job performed on the successful implementation of Cash IVR.                         Commercial

我运行以下代码来获取主题

x <- udpipe(x= praise$Feedback.Comments, object= ud_model)

x$topic_level_id <- unique_identifier(x, fields = c("doc_id", "paragraph_id", "sentence_id"))
## Get a data.frame with 1 row per id/lemma
dtf <- subset(x, upos %in% c("NOUN", "ADJ"))
dtf <- document_term_frequencies(dtf, document = "topic_level_id", term = "lemma")
head(dtf)


dtm <- document_term_matrix(x = dtf)
## Remove words which do not occur that much
dtm_clean <- dtm_remove_lowfreq(dtm, minfreq = 5)
head(dtm_colsums(dtm_clean))

#dtm_clean <- dtm_remove_terms(dtm_clean, terms = c("%age", "4G"))
## Or keep of these nouns the top 50 based on mean term-frequency-inverse document frequency
dtm_clean <- dtm_remove_tfidf(dtm_clean, top = 80)

library(topicmodels)
m <- LDA(dtm_clean, k = 4, method = "Gibbs", 
         control = list(nstart = 5, burnin = 2000, best = TRUE, seed = 1:5))

scores <- predict(m, newdata = dtm, type = "topics", 
                  labels = c("labela", "labelb", "labelc", "xyz"))
str(scores)

predict(m, type = "terms", min_posterior = 0.05, min_terms = 3)


dtf <- subset(x, upos %in% c("NOUN", "ADJ"))

dtf <- document_term_frequencies(dtf, document = "topic_level_id", term = "lemma")
dtm <- document_term_matrix(x = dtf)
dtm_clean <- dtm_remove_lowfreq(dtm, minfreq = 3)
## Build topic model + get topic terminology
m <- LDA(dtm_clean, k = 4, method = "Gibbs", 
         control = list(nstart = 5, burnin = 2000, best = TRUE, seed = 1:5))
topicterminology <- predict(m, type = "terms", min_posterior = 0.025, min_terms = 5)
scores <- predict(m, newdata = dtm, type = "topics")

我得到的结果如下

$topic_001
             term       prob
1            work 0.31616890
2        customer 0.13422588
3            time 0.08616547
4            role 0.05526948
5   collaborative 0.03810505
6         service 0.03810505
7           value 0.03810505
8         amazing 0.03123927
9  implementation 0.03123927
10           line 0.02780639

我想通过Division_name得到它们中的每一个

我想要的结果

             term       prob        Division
1            work 0.31616890        Finance
2        customer 0.13422588        Finance
3            time 0.08616547        Commercial
4            role 0.05526948        Commercial
5   collaborative 0.03810505        Commercial

模拟数据集

structure(list(Feedback.Comments = c("Excellent kick start of p", 
"Nauman is very collaborative when it comes to team deliverable. He takes ownership and ensure to support whenever needed or asked for. ", 
"Thank you for being very collaborative and designing and planning the whole workshop that deemed success today for R", 
"Amazing knowledge sharing session conducted by you. Truly innovative.", 
"Thanks a lot for your collaboration during my training dates", 
"During Prepaid Consolidation Step 1, you have done excellent job in handling the Mediation stream resulting in a smooth delivery.  The highlights of this delivery was the collaboration which was executed excellently.", 
"He handles all the organization customers in a very collaborative manner.", 
"Noor ul Amin is very supportive and initiative hungry person, always take very quick/bold step when ever any issue happened. ", 
"Keeping check on timely rectification of observations by HSSE with good speed.", 
"Smooth execution of Regional Administration in my absence. Well done.", 
"Good Job performed on the successful implementation of Jazz Cash IVR. the 1st selfservice IVR for financial transactions in industry.", 
"Despite challenges on the resource side you have done exceptionally well in managing the UATs, Prepaid Consolidation and assigned tasks.\n\nWe need focus more on FCR & NPS related areas so we are able to meet our KPIs, looking forward for stats and feedback on time. It would be better if we dedicate one resources on this side and not deploy all resources on prepaid consolidation (it will not give us any benefit)", 
"Job well done in reorganizing all the investments to fixed portfolios.\n Keep it up.", 
"Well done in reorganizing PF process and resolving legacy issues.", 
"Job well done in completing CPs and making the facility available well in time.", 
"You always seems supportive on these requests.sometimes you also submitted input in late hours of the day. Keep ot up.", 
"Well done on completing Hiperos screening for almost 30 profiles. Please pass the feedback to Khurram and Babar.", 
"You always make your concerns clear at a judgment. It always good to have a critical view on things, helps avoiding mistakes. Keep it up.", 
"Both FLT in Lahore and Karachi were planned, managed and executed to the perfection under your lead. Wonderful collaboration with P&O and cross functional teams. Good job and good management. ", 
"Very good resource. Always up to the expectations.\nDid good job in back office evaluations"
), division_name = c("People & Organization", "People & Organization", 
"People & Organization", "People & Organization", "People & Organization", 
"Technology", "Finance", "Finance", "People & Organization", 
"People & Organization", "Commercial", "Commercial", "Finance", 
"Finance", "Finance", "Finance", "Finance", "Finance", "People & Organization", 
"Commercial")), row.names = c(1L, 3L, 4L, 5L, 6L, 7L, 8L, 9L, 
10L, 11L, 17L, 21L, 23L, 24L, 25L, 29L, 31L, 32L, 35L, 37L), class = "data.frame")

标签： rldatopic-modelingudpipe